Trunk Recorder Pro 0.1.0

The first release of the Rust rewrite: one self-contained program per
platform, with no GNU Radio, OP25 or other runtime dependencies.

• P25 Phase 1 control channel and voice (C4FM and CQPSK/LSM simulcast),
  with receiver diversity and soft-decision FEC: 98–99 % of control messages
  on a simulcast site where op25's decoder gets 62 %.
• P25 Phase 2 TDMA voice (H-DQPSK, AMBE+2), descrambled with the
  system's WACN / System ID / NAC from the control channel.
• Motorola SmartNet / SmartZone (Type II) control channels: 3600 baud
  2FSK, soft-Viterbi OSW decoding, Trunk Recorder's OSW parser and band
  plans (800 standard / rebanded / splinter, 900, OBT 400_custom). Voice
  follows the grant: P25 Phase 1 for digital grants, narrowband FM for analog
  (defaultMode for talkgroups never heard granted). Configured with Trunk
  Recorder's names (type "smartnet", bandplan, bandplanBase, …). On
  WMATA's OBT system: 90–99 % of OSWs where Trunk Recorder gets 40–60 %.
  SmartNet systems work alongside P25 ones (each system picks its type), and
  Trunk Recorder configs with SmartNet systems import.
• DMR: conventional channels (mode "dmr", each slot its own calls,
  with the colour code) and trunked sites (type "dmr"): Capacity Plus and
  Linked Capacity Plus followed by their link control on every repeater,
  Capacity Max / Connect Plus / Tier III by their grants. Logical channel
  frequencies are learned from the air (or Trunk Recorder's lcnTable);
  keyed checksums (restricted access) are recognised. Repeated blocks and
  embedded link control are soft-combined, which recovers most of them on a
  weak site. Voice through the Phase 2 AMBE+2 decoder. Trunk Recorder DMR
  configs import. tool dmrscan / tool dmr for captures.
• DMR in Find my system and the dashboard: the scan recognises trunked
  DMR sites (kind and colour code) and adds them; Business UHF / VHF bands.
  The dashboard shows a DMR site's rest channel, watched frequencies with
  each slot's call, and its channel table. Simplex / talkaround DMR (a
  mobile's bursts, nothing between them) is received: the 4FSK receiver
  leaves quiet stretches out of its timing and levels.
• Weak signals: the C4FM receiver (P25 C4FM and DMR) has a matched
  filter, takes its levels from the symbol clusters, and re-decides
  symbols by multi-symbol detection below the discriminator's threshold;
  SmartNet decides by tone energies. Half the codewords decoded at about
  6 dB less signal on P25 C4FM voice, 8 dB on DMR voice, 1.7 dB on
  SmartNet (tool snr, which measures it; see research/weak-signal.md).
• Find my system finds SmartNet too, and learns its band plan from the
  air (which carrier comes up when a channel number is granted) — on WMATA,
  exactly Trunk Recorder's hand-made 400_custom plan, all four control
  channels and the voice channels, from one dongle.
• Talker aliases: the names radios send over the air with their unit
  ID (Motorola and Harris, Phase 1 and Phase 2 — Trunk Recorder's
  decoders), on trunked systems (SmartNet's P25 voice included) and
  conventional P25 channels. On DC's system: the same names Trunk Recorder
  decodes. The interface shows a radio's alias in place of its unit ID
  (the ID on hover), in live calls and history, and history can be
  filtered by it. Encrypted calls aren't recorded, but their voice channel
  is still followed (when a recorder is free) for the link control sent in
  the clear: who spoke and their aliases. Each system's aliases are kept between runs in
  <shortName>.units.csv, Trunk Recorder's unitTagsOTA format, and saved
  in each call's srcList as tag_ota.
• Conventional channels: analog narrowband FM and P25, alongside a
  trunked system or on their own. Found by energy in the spectrum the
  channelizer already computes (an idle channel costs almost nothing), opened
  with pre-roll so transmissions start in full; per-channel mode, name,
  talkgroup and squelch (dB above the measured noise floor). Kept in the
  interface (with CSV import and export), in the config, or in a linked CSV
  channel file to edit in a spreadsheet; Trunk Recorder's channel file and
  config import.
• Unit IDs on analog calls: MDC1200 and FleetSync (1200 and 2400 baud)
  bursts are decoded from the audio of analog calls, conventional and
  SmartNet, into srcList; an MDC1200 emergency flags the call. Always on,
  at about 0.02 % of a core per analog call being recorded.
• Tones on analog calls: the CTCSS tone (all 51) or DCS code (all 112,
  either polarity) each conventional FM call carries is identified and
  written to the call JSON as Trunk Recorder's tone_detected /
  tone_confidence, and shown in the call list.
• Tones on conventional channels: an analog channel's Tone (typed as
  RadioReference or Trunk Recorder write it: 151.4 PL, 023 DPL, D023N)
  records only transmissions carrying it. Several rows on one frequency
  split it by tone, each filed under its own talkgroup (numbered for you);
  a row with no tone takes the rest. Transmissions are held until their tone
  is known, so nothing is cut off. Trunk Recorder's channel file Tone
  column reads as is.
• NAC and colour code on conventional channels: the same Tone column
  takes a P25 NAC (293 NAC, $293) or DMR colour code, slot and
  talkgroup (RadioReference's CC1 TS2 TG201), so P25 users of one
  frequency are told apart by NAC and a DMR repeater's traffic picked by
  slot and talkgroup.
• Heard codes: under each conventional frequency, the tones, NACs and
  colour codes it has carried (recorded calls and transmissions no row
  took), kept between runs, each with an Add that makes a row for it.
• Find my system: a first-run survey for people who don't know their
  frequencies. It scans the land-mobile bands (800 / 700 / 900 MHz, UHF,
  VHF, optionally UHF federal and T-band) for carriers that never key down,
  checks each with the P25 receivers, then listens to the best control
  channel and reports its WACN / System ID / NAC / site, band plan,
  alternate control channels, neighbouring sites and voice channels. From
  the frequency the channel announces versus where it is heard, it measures
  the radio's frequency correction (ppm); on an RTL-SDR it also picks the
  gain. Add this system adds it with a site lock, and sets the ppm, gain
  and a center that covers the most voice channels seen. Also trunk-pro survey.
• Several systems and sites at once: each P25 system — or each site of
  a multi-site system — follows its own control channel, with its own short
  name (folder), band plan, talkgroups and modulation, sharing the sources
  and recorders. A site lock (NAC, WACN, System ID, RFSS, site; filled in
  by the survey) keeps each on its own control channel; neighbouring sites a
  control channel announces can be added with one click. The dashboard
  groups sites of one system and filters calls, history, log and live audio
  by system; sources left on Auto are placed over the systems not yet
  covered. Configs with one system carry over unchanged, and a Trunk
  Recorder config's P25 systems are all imported. trunk-pro replay --system name:Hz[:nac=…,site=…].
• Calls heard on several sites are saved once (Trunk Recorder's
  multiSite). Sites are grouped into systems automatically, by the WACN /
  System ID (P25) or System ID (SmartNet) their control channels announce;
  a siteGroup name groups DMR sites or ISSI-linked systems. Where Trunk
  Recorder records only the first site's grant, every copy is recorded here,
  and the one decoded most cleanly is kept, so a site fading mid-call
  doesn't cost the call. The talkgroup CSV's Preferred Site (or Trunk
  Recorder's Preferred NAC) wins when its copy is nearly as good. Live
  audio plays one copy; the dashboard shows "also on …". On by default
  (recording.dropDuplicateCalls). Trunk Recorder imports keep their
  multi-site settings. --system …:group=name for replays.
• Several RTL-SDRs feeding the systems, over a pure-Rust USB driver (no
  librtlsdr / libusb to install). New dongles start at 25.4 dB gain: higher
  gains overload the front end near strong 800 MHz transmitters (about 3x
  the voice-frame errors at 38.6 dB).
• USRP and Airspy sources (optional): used when UHD / libairspy is
  installed, found at run time by the same binary.
• Captures in cu8, cs16 or cf32 (GNU Radio / UHD) formats.
• Trunk Recorder–compatible output: WAV + call JSON in Trunk Recorder's
  folder layout and field names; talkgroup CSV import; import of a Trunk
  Recorder config. error_count is the call's FEC-corrected bit errors,
  and errorList breaks them down per 10 s of audio (errors, repeated or
  muted frames, worst frame) so bursts stand out.
• Plugins: programs of their own that the recorder runs while it records
  and tells what happens — calls starting, ending and landing on disk, radio
  activity, live audio, status — over JSON lines on stdin/stdout. They only
  watch; a plugin that crashes is restarted, and one that falls behind loses
  events rather than slowing the recorder. Settings in plugins.json
  beside the config. Calls are encoded to M4A once for every plugin that
  asks (ffmpeg, macOS's afconvert or fdkaac, whichever is there; WAV only
  without one). trunk-pro plugin list | describe | run to look at plugins
  and run one against recorded calls. The trunk-recorder-plugin crate is
  the protocol and a Rust SDK. A Plugins page lists them: turn each on
  or off, fill in settings drawn from what the plugin describes (its own and
  for each system), see whether it's running and how many calls it handled,
  its recent log, and the M4A encoder. Changes apply at once while recording.
• Vocoder frame capture (setting "Save vocoder frames"): each call's
  decoded voice frames and error counts as <call>.frames.jsonl;
  trunk-pro tool revoice vocod...