Make DicomTile objects more accessible

[melissalinkert]

Adds a bit of API to DicomReader to make it easier to get DicomTile objects corresponding to a particular resolution and/or plane. This also improves indexing of tiles, which should hopefully make image reading faster overall.

Local tests pass, but will exclude this temporarily if tests fail with it included. Since this is expected to improve performance, will need to compare total runtime of passing tests with and without this change, for several days each.

[sbesson]

This PR introduces a couple of public APIs allowing to access the internal tiles associated with each series and precomputes the index of plane associated to every single tile.

Looking at the timing of the integration tests

for i in $(seq 1 30); do curl -s -o log https://bf-testing-results.s3.amazonaws.com/2026/2026-06-$(printf %02d i)/dicom.log && grep "Total time:" log; done 
for i in $(seq 1 30); do curl -s -o log https://bf-testing-results.s3.amazonaws.com/2026/2026-07-$(printf %02d i)/dicom.log && grep "Total time:" log; done 

the average completion time is 260 min +/- 14 min between 2026-06-01 and 2026-07-03 (before this PR was introduced) vs 249 min +/- 18 min between 2026-07-04 and 2026-07-30.

Is it correct that the major performance improvements will be seen when calling openBytes or openCompressedBytes as the tile lookup should benefit from the preliminary computation? In that case, is it pertinent to also measure the timing of a conversion from DICOM to OME-TIFF for some sample WSI DICOM?

[sbesson]

Should this be getZOffsets(getCoreIndex()) to use the new API?

[melissalinkert]

See c3997d6

[melissalinkert]

Is it correct that the major performance improvements will be seen when calling openBytes or openCompressedBytes as the tile lookup should benefit from the preliminary computation? In that case, is it pertinent to also measure the timing of a conversion from DICOM to OME-TIFF for some sample WSI DICOM?

Yes, I'd expect openBytes/openCompressedBytes to be a bit faster overall, so converting is probably the easiest way to iterate over all tiles to test that assumption.

[sbesson]

I ran bfconvert -noflat and bfconvert -noflat -precompressed with and without this PR using https://downloads.openmicroscopy.org/images/DICOM/wsi/2023-04-28/PARNED-0BNNF4_B2_Q30/ as the input using 3 repeats.

Average conversion times are the following:

	Bio-Formats 8.5.0	PR 4451
bfconvert -noflat	559s +/- 12s	548s +/- 9s
bfconvert -noflat -precompressed	143s +/- 3s	139s +/- 2s

In summary, this PR is improving access to the internal DICOM tiles via a new public API and refactors a few places in the reader. The nightly repository tests confirm the reading is unaffected and the conversion tests confirm no regression and a possible slight performance improvement. Were you expecting a more substantial writing change and would you like me to perform additional conversion testing on another sample?

Otherwise, no objections to merging from my perspective

[melissalinkert]

Were you expecting a more substantial writing change and would you like me to perform additional conversion testing on another sample?

I didn't have a specific performance change in mind, other than "not worse, maybe slightly better". I think the test you have done is sufficient to confirm that.