Add return_indices to local_thickness and porosimetry for compact output

[ma-sadeghi]

Closes #840.

The default float64 output of local_thickness/porosimetry becomes the peak-memory bottleneck on large tomograms (~21 GB for a 1500x1500x1200 volume in OP's case). With return_indices=True these now return (sizes, indices) where indices is uint8/uint16 and sizes[indices] recovers the float result. Output memory drops 8x or more.

To save memory inside porosimetry too (not just at the API boundary), drainage_dt gained a matching return_indices flag that skips allocating the float im_size array and writes the step index into a compact im_seq. The mode is opt-in and not compatible with outlets, since trapping post-processing relies on im_size.

Only method='dt' is supported for now; the other methods raise NotImplementedError. Easy to extend to conv/bf in a follow-up if needed.

# Before: 21 GB float64 output
lt = ps.filters.local_thickness(im, sizes=25)

# After: ~3 GB uint8 output, sizes[indices] recovers the float
sizes, indices = ps.filters.local_thickness(im, sizes=25, return_indices=True)

Default behaviour is unchanged. Two new tests added; full unit suite green.

[codecov]

Codecov Report

❌ Patch coverage is 84.61538% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 70.5%. Comparing base (c352d60) to head (5c6468c).
⚠️ Report is 62 commits behind head on dev.

Additional details and impacted files

@@          Coverage Diff          @@
##             dev   #1164   +/-   ##
=====================================
  Coverage   70.4%   70.5%           
=====================================
  Files         45      45           
  Lines       6460    6488   +28     
=====================================
+ Hits        4554    4576   +22     
- Misses      1906    1912    +6     

🚀 New features to boost your workflow:

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

[ma-sadeghi]

The test_bond_number failure on py3.13 ubuntu is the same flake fixed in #1161 — it'll clear once that lands and this branch is rebased. Everything else is green.

[jgostick]

I wonder if this is the right way to fix this? The reality is that all spheres drawn in a digital image have integer radii, so writing fractional radii is actually an artifact of the way the sizes are selected. If sizes is given as N, then it pulls out N sizes between 1 and max(dt). We could change the default behavior of the size generator, and then make the output (im_sizes) respect the dtype of the sizes? (i.e. int32)