Add ICD coverage for channel and stokes changes on an inactive file

[markccchiang]

Description

Closes #31. Adds ICD coverage for SET_IMAGE_CHANNELS sent against an image the user is not currently working on, the protocol behaviour behind carta-frontend #2206.

That PR made the frontend update the channel and the stokes of every frame rather than only the active one, and it sends a different message shape depending on whether the frame is being rendered:

frame	frontend call	message
visible	TileService.requestTiles(...)	SET_IMAGE_CHANNELS with a full required_tiles
not visible	TileService.updateHiddenFileChannels() → setChannels(fileId, ch, st, {})	SET_IMAGE_CHANNELS with an empty required_tiles

The empty required_tiles shape was untested. ANIMATOR_NAVIGATION already covers two files changing channel and stokes with tiles, so the genuine gaps were: the empty message, the back-to-back pair, and per-file state isolation.

What the test pins

Session::OnSetImageChannels sets channel/stokes, runs UpdateImageData, then only forwards to OnAddRequiredTiles if (message.has_required_tiles()), which itself early-returns on an empty tile list. So the contract for a hidden frame is: histogram yes, tiles no, and the state still moves.

1. Empty-tiles update on file 1 -> REGION_HISTOGRAM_DATA at the new channel/stokes, and no RASTER_TILE_SYNC or RASTER_TILE_DATA at all.

2. File 0 re-reads tiles -> still at its own channel 0 / stokes 0, unaffected.

3. File 1 re-reads tiles -> serves at channel 2 / stokes 1. No further SET_IMAGE_CHANNELS was sent after step 1, so that position can only have come from the empty update.

4. Both files updated back to back without waiting -> two complete, correctly tagged tile groups.

Step 4 asserts per file_id rather than by order of arrival. The backend currently serialises the two handlers under _frame_mutex, but that is an implementation detail, not the interface.

Changes

File
src/test/TILE_DATA_INACTIVE_FILE.test.ts	new - 27 tests
ICD_test_stages/raster_tiles.tests	register the test in the raster-tiles stage
docs/source/tile_data_request.rst	document the 10 steps and Checks 1-6

Fixtures are M17_SWex.hdf5 (active) and HH211_IQU.hdf5 (inactive). HH211_IQU carries a spectral and a Stokes axis, so channel and stokes can be changed together.

Two local helpers

Stream(CARTA.RasterTileData, n) takes two messages off rasterSyncStream and n-2 off rasterTileStream, resolving on the end sync. That assumes a single sync group, and it hangs rather than fails when nothing arrives, so it can express neither of the two things this test needs:

• RasterStream(syncCount, tileCount) counts the two subjects separately, so it spans the two sync groups produced by step 4.

• RasterSilence(durationMs) collects whatever arrives in a fixed window and resolves regardless, which is what makes the negative assertion in step 1 expressible.

Checklist

For the pull request:

• Documentation has been updated (or no documentation changes are needed)

[loveluthien]

😎