v1.0.6

Sweep now captures packets while it scans. Previously, running Sweep in
repeat mode meant the receiver heard essentially nothing — measured at 0 of
42 real packets during a 4-minute run. Repeat Sweep now pauses on your home
channel between laps to actually receive, and captured packets appear in
detections.csv, the Captures card and the Nodes roster exactly as if Trace
had caught them. Measured 30 of 44 packets (68%) over two runs against a
real mesh.

New setting — System > Tuning > Capture (Off / 1s / 2s / 4s,
default 2s). This is the trade: a longer window captures more packets, a
shorter one gets you round the band faster. At the 2s default a full-band
survey takes about 2.9s instead of 0.9s. Set it to Off for the old
scan-only behaviour. Saved to /loratrace/capture.txt.

Waterfall now distinguishes what it heard from what it decoded. Green
marks a bin where a real packet was demodulated and CRC-checked; yellow
remains an energy reading above the margin. Green often appears with no
yellow beneath it — that is correct, not a glitch: Sweep's per-bin look is
milliseconds against a packet lasting 142-490ms, so it frequently misses
traffic the receiver then decodes cleanly. The header reads N PKTS in
green when packets were captured but no energy peak was found, instead of
the flat QUIET it used to show while traffic was actively being recorded.

Fixes

• Starting a Probe or Cell scan during a repeat Sweep is now correctly
  refused instead of being silently queued and firing minutes later.
  Stopping repeat Sweep also takes effect immediately rather than waiting
  out the capture window.
• The Waterfall could show a green packet mark on a sweep that had received
  nothing, if repeat mode was stopped mid-window. Those readings are now
  discarded.
• Adjusting System > Tuning > Margin mid-sweep no longer judges one lap
  against two different thresholds.
• A corrupt or hand-edited capture.txt or display.txt could silently
  select "Off" for the capture window or idle-dim, because unparseable text
  was being read as 0. Bad values are now ignored and the previous setting
  is kept.
• Sweep's radio ownership, the channel shown on status pages, and the data
  behind the Waterfall are all hardened against cross-core races. No
  symptom was reported for these; they were found by audit.

Known issues

• energy.csv's absolute RSSI values read about 2.4 dB lower than they did
  before v1.0.2. Peak detection and the 35 dB margin are unaffected — only
  comparisons of raw RSSI across that boundary. See
  docs/LOG_GUIDE.md.
• The GitHub Pages web flasher previously failed to update when a release
  was published. A fix is included but is unverified until a release is
  actually published with it in place.