smartspim: prefer SmartSPIM's stitched All_Channels volume

[stevevanhooser]

Summary

ferret-lightsheet (formats/SmartSPIM.md, spim2zarr --source all_channels) established that All_Channels/ contains SmartSPIM AutoStitch's TeraStitcher merge of the destriped tiles, with flat-field and left/right illumination correction and blended overlaps. This is the better-stitched, harmonized data that Imaris is built from. With this PR, ndr.reader.smartspim reads it first. The PR also fixes two TIFF data-type bugs found along the way (see below).

All_Channels layout:

• A flat folder of 2-D LZW-compressed uint16 TIFFs, one per (z, channel), named <Y>_<X>_<z*10>_ch<k>.tiff, plus an ImageList.txt.
• Each slice covers the whole stitched canvas, so there are no tiles.
• ch<k> follows the filter-channel number, not the acquisition order. It comes from <channel>_MIP_stitched/, or from the filter-channel rank when that folder is missing.

New format functions (+ndr/+format/+smartspim)

• readAllChannelsInfo: indexes the slices (sorted by numeric z) and maps each channel to its ch<k>. Also returns the canvas size and class from the first TIFF header, and the voxel size from metadata.json.
• hasAllChannels: a non-throwing probe that checks file names only. It returns true only when every channel has slices.
• readAllChannelsVolume: reads one channel as (z, y, x), with an optional ZRange.

Reader epochs

Epochstream	Reads
{root}	stitched volume, all channels on the C axis (ch<k> order)
{root, channel}	stitched volume, one channel
{root, channel, tile}	destriped Ex_* tile (unchanged)

Fallback: the Ex_* tiles are not obsolete; ferret-lightsheet treats them as the source of truth. When All_Channels is missing or incomplete, {root} and {root, channel} raise the same ChannelNotPinned / TileNotPinned errors as before.

New resolveepoch fields: source, channelNames, numSlices, height, width, dtype and allChannels. info.tile is unchanged for tile epochs.

Fix 1: readTileInfo reported int16 for some uint16 TIFFs (commit 3e300e5)

• Cause: tiffDtype treated any SampleFormat containing "signed integer" as signed. imfinfo's 'Unsigned integer' contains that text.
• Effect: a TIFF that stores SampleFormat = 1 explicitly came back as int16. readframes then allocated an int16 array, so every pixel above 32767 was clipped.
• Why it hadn't shown up: tifffile (which writes the destriped tiles) and MATLAB's imwrite don't store the tag. libtiff-based writers may.
• Fix: rule out 'unsigned' before testing for 'signed'.
• Tests: new tests write TIFFs with an explicit UInt / Int SampleFormat tag and check readTileInfo and readAllChannelsInfo. readAllChannelsInfo already used the corrected logic.

Fix 2: tiffstack.tiffclass read float TIFFs as uint32 (commit 8f870c1)

• Cause: tiffclass matched SampleFormat by exact string, and its float case accepted only 'ieeefloat' / 'float'. imfinfo reports float TIFFs as 'IEEE floating point', so they fell through to the unsigned branch.
• Effect: a float32 stack came back as uint32. readframes then cast every value to uint32, dropping fractions and clamping negative values to 0.
• Fix: match by substring, with the same unsigned-before-signed order as Fix 1.
• Scope: this also covers ndr.reader.prairieview, which reuses tiffclass.
• Tests: TestTiffstack now round-trips a float32 stack and an int16 stack. writeMultipageTiff takes an optional SampleFormat; existing callers still get uint16.

Tests

• makeExampleFixture(..., 'AllChannels', true) adds an LZW All_Channels folder to the test fixture.
• New test classes: TestSmartSPIMAllChannels and TestSmartSPIMReaderAllChannels.
• The existing tests (fixture without All_Channels) still cover the fallback path.
• CI at 3e300e5: 396 tests, 0 failures. The tiffstack tests were added afterwards in 8f870c1.

The Python port is VH-Lab/NDR-python#31. Its bridge YAML records b8f0e4b (All_Channels), 3e300e5 (readTileInfo) and 8f870c1 (tiffstack). The Python readers get the data type directly from tifffile and never had either bug.

🤖 Generated with Claude Code

https://claude.ai/code/session_01U33UPxYJWPpcQimNu5t5Cb

[github-actions]

Test Results

398 tests  +32   395 ✅ +32   36s ⏱️ ±0s
 51 suites + 2     3 💤 ± 0 
  1 files   ± 0     0 ❌ ± 0 

Results for commit 8f870c1. ± Comparison against base commit 95b7e2d.

♻️ This comment has been updated with latest results.

[codecov]

Codecov Report

❌ Patch coverage is 91.15385% with 23 lines in your changes missing coverage. Please review.
✅ Project coverage is 54.26%. Comparing base (95b7e2d) to head (8f870c1).

Files with missing lines	Patch %	Lines
+ndr/+format/+smartspim/readAllChannelsInfo.m	87.27%	14 Missing ⚠️
+ndr/+format/+smartspim/readAllChannelsVolume.m	78.04%	9 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #142      +/-   ##
==========================================
+ Coverage   53.26%   54.26%   +1.00%     
==========================================
  Files         160      163       +3     
  Lines        9190     9420     +230     
==========================================
+ Hits         4895     5112     +217     
- Misses       4295     4308      +13     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.