…d-transfer caller error

PR #370 review feedback (jrusso1020):

- chunkEncoder: when codec=h264 and hdr is set, log a warning and strip
  hdr instead of emitting a half-HDR file (BT.2020 container tags +
  BT.709 VUI inside the bitstream). libx264 has no HDR support; the only
  honest output is SDR/BT.709. Caller is told to use codec=h265.

- videoFrameExtractor: comment at the convertSdrToHdr call site clarifying
  that dominantTransfer is majority-wins; mixing PQ and HLG sources in a
  single composition is caller-error and the minority transfer's videos
  will be converted with the wrong curve. Render two compositions if you
  need both transfers.

- docs/guides/hdr.mdx: limitations section now documents (a) H.264 + HDR
  is rejected at the encoder layer, and (b) GPU H.265 (nvenc, videotoolbox,
  qsv, vaapi) emits BT.2020 + transfer tags but does NOT embed master-display
  or max-cll SEI, since ffmpeg won't pass x265-params through hardware
  encoders. Acceptable for previews, not for HDR10 delivery.

…for all shaders

Existing smoke tests cover only the endpoints (p=0 ≈ from, p=1 ≈ to),
which miss a class of regressions that surface specifically at the
midpoint — where the transition is most visible to viewers — and let
shaders silently rot in CI:

  • A shader becomes a no-op (returns input as-is)
  • A shader prematurely completes (returns target at midpoint)
  • A shader doesn't write to the output buffer at all
  • A shader loses determinism (Math.random / Date.now / leaked state)

Add four invariants every shader must satisfy at p=0.5, applied via a
describe loop over ALL_SHADERS so any new transition added to the
registry automatically picks up the same coverage:

  1. output ≠ from           catches no-ops
  2. output ≠ to             catches premature completion
  3. output is non-zero      catches blank output
  4. output is deterministic catches accidental non-determinism

Uses two distinct uniform input colors (40000/30000/20000 vs
10000/10000/10000) so equality checks have distinct byte patterns to
compare against. Even shaders that warp UVs (which would be no-ops on
uniform input alone) produce mix16(from, to, 0.5) at every pixel,
distinct from both inputs.

60 new tests (4 invariants × 15 shaders), all passing.

Follow-up to plans/hdr-followups.md Chunk 9G.