add feature extractor TICON and bug fixes

[mducducd]

• Add new h_optimus1+ticon extractor. In the mean time, this TICON checkpoint achived best benchmark with H-Optimus-1 as the tile encoder whether it's from individual or omni pretraining; a separate feature-level TICON stage can be considered later as more state-of-the-art foundation extractors become available. Also they will public tangle model for slide-level later.
• Fix: TITAN operates on patch grid geometry, not microns, so patch spacing must be inferred from coordinate stride rather than assumed from physical size.
• Fix: Correcting StratifiedKFold by event/status (survival analysis)

[FWao]

@mducducd Please add tests.

We should also do some end-to-end tests to validate TICON works.

[mducducd]

feature_extractor test takes all extractors in ExtractorName collection.

STAMP/src/stamp/preprocessing/config.py

Line 14 in 16d38d1

class ExtractorName(StrEnum):

I rerun extractor test and all 17 extractors was tested including Ticon (h_optimus_1 integrated). I am not sure the run checks showed on github, virchow family are also missing.

[s1787956]

please add some more comments on what is happening here

[s1787956]

please reference the preprint and github repo where we adapted this from

[s1787956]

We maybe should / need to do this for all encoders / models that we use / adapt from.. they usually also come with licenses.. but I think a proper reference should do it for now. Would be best at beginning of the specific file (additionally summarized at end of readme or getting started guide)