v1.27.0

• The conformed clip keeps its SOURCE version current (Chat 99, measured
  in-vivo — the fix for a full day of 'no media' on the timeline): Flame read
  the very same .clip as start_frame=1001 when the light version was
  current and as start_frame=0, spanning 1101 frames, when the comp version
  was. With the comp current, every Update Sources — which performs a
  replace — anchored the conformed segment at 00:00:00:00 and lost its
  cut. The asymmetry is what cost the time: Update Sources BREAKS the
  anchor against a badly-read clip and does not restore it against a good
  one (verified across all three version_selection_mode values), so a
  damaged segment must be re-laid with PySequence.overwrite. New
  keep_source_current marks the first (source) version current instead of
  the newest; the comp version is still in the clip and the operator flips to
  it natively, which is the intended gesture. Nothing is written to the
  timeline. Validated end to end: re-conform → update_sources() → flip to
  comp → flip back, anchor 00:00:40:01 throughout.

• Spliced feeds inherit the source feed's anchor DESCRIPTION, not just its
  ticks: the two feeds already agreed on nbTicks 1001 and the clip still
  resolved to 0, because they disagreed about how that timecode is expressed
  — Maya's EXRs carry no timecode (TimecodeSource=Filename, empty rate, no
  sampleRate) while Flame's Write File embeds one (Header, explicit
  rate). startTimecode/rate, sampleRate and the TimecodeSource /
  RateSource userData entries are now copied from the source feed, ALWAYS —
  the previous early-return skipped it whenever the ticks matched. Pixel
  metadata (channels, layers, colour space) is deliberately left alone. This
  alone did NOT fix the start-frame reading; it is kept because leaving the
  feeds inconsistent about their own anchor is a latent trap. +7 tests.

• openclip_create can splice in publishes selected by TYPE, with no Task
  (Chat 99): Flame's own tk-flame batch-render integration publishes the comp
  as Flame Render and links no Task — its context comes from the
  .batch path, whose template carries no {Step} token. Every existing
  selector is Task-based, so the conformed clip never learned the comp
  existed and Update Sources had nothing to offer. New optional
  extra_publish_types (e.g. ['Flame Render']) adds one version group per
  type, matched on the Shot alone, appended AFTER the Task-selected groups so
  the comp lands on top and becomes current. The uid token is read from the
  publish code (SEQ003_SH001_CMP_v001.%04d.exr → CMP_v001) rather than
  from the type name, so the clip reads LIGHT_v003 + CMP_v001. A type
  with no publishes yet is reported in skipped, never fatal — the clip is
  valid before the first comp render. Still explicit by design: the tool does
  not guess which publishes feed a conform. +5 tests.

• Spliced versions inherit the SOURCE version's start-timecode anchor
  (Chat 99, in-vivo 'no media' on the timeline flip): a conformed segment
  lines its versions up by TIMECODE, and the two writers disagree — Maya's
  EXRs carry no timeCode attribute so dl_get_media_info falls back to
  the frame number (LIGHT feed nbTicks 1001), while Flame's Write File
  embeds a real one and a batch whose source timecode was never set stamps
  00:00:00:00 (COMP feed nbTicks 0). Same frame numbering, same
  duration, correct paths — and the flip still showed 'no media', because
  the segment asked for TC 1001-1100 against a feed spanning TC 0-99. The
  pipeline owns the clip, so splice_openclips now pulls every later
  feed's nbTicks/dropMode onto the first (source) version's, per track,
  and openclip_create REPORTS it in a new timecode_realigned field —
  never a silent metadata rewrite. startFrame is deliberately NOT
  normalised: it must keep matching the real filenames in <path>, and
  numbering parity is enforced upstream at render time. +6 tests.

• openclip_create multi-step aggregation (Chat 98 comp architecture,
  operator-approved): new optional steps list splices publishes from
  SEVERAL Steps into ONE clip, in list order — ['Light', 'Comp'] gives
  the conform timeline the LGT render and every comp version through the
  same open clip, flipping natively via Source Versions. Version uids are
  disambiguated with the step string uppercased (LIGHT_v003,
  COMP_v001); current = the newest version of the LAST listed step with
  publishes; a listed step with no publishes yet is reported, never an
  error (the clip is valid before the first comp render). Proven necessary
  in-vivo: Flame's Write File OWNS any clip it creates — it overwrote the
  pipeline clip wholesale (43 KB multichannel → 2.4 KB, versions gone), so
  the pipeline aggregates and the Write File is demoted to media-only.
  Single-step behaviour unchanged (plain vNNN uids). +3 tests.