Xenium: cache zip store opening; vectorize cell id encoding

[LucaMarconato]

After doing some profiling, I identified 2 bottlenecks:

1. the zipstore was opened multiple time and not reused.
2. the cell_id_str_from_prefix_suffix_uint32() had some long for loops that could be removed.

This PR:

1. caches the opening of the cells zip store
2. vectorizes the function cell_id_str_from_prefix_suffix_uint32()

The second part is very interesting: Claude Opus 4.6 suggested a clever bit-shift trick which halves execution time (illustrated below, again with the help of Claude).

Further optimization could be explored for the string concatenation, which is now driving the runtime of the function (10% of total, see flame graph), and is among the top 3 drivers of runtime.

Flame graph:

Explanation of the bit shift trick:

[codecov-commenter]

Codecov Report

❌ Patch coverage is 93.44262% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 62.86%. Comparing base (7eb2e61) to head (563693a).
⚠️ Report is 22 commits behind head on main.

Files with missing lines	Patch %	Lines
src/spatialdata_io/readers/xenium.py	93.44%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #370      +/-   ##
==========================================
+ Coverage   62.70%   62.86%   +0.15%     
==========================================
  Files          27       27              
  Lines        3140     3156      +16     
==========================================
+ Hits         1969     1984      +15     
- Misses       1171     1172       +1     

Files with missing lines	Coverage Δ
src/spatialdata_io/readers/xenium.py	72.47% <93.44%> (+0.96%)	⬆️

🚀 New features to boost your workflow:

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

[LucaMarconato]

• The latest commit removes some unnecessary conversion to string for some dataframe indices.

Flamegraph below shows the current status. The string operations in cell_id_str_from_prefix_suffix_uint32() definitely consume quite some time.

[LucaMarconato]

• cell_id_str_from_prefix_suffix_uint32() was being called twice, now this is avoided. Old versions don't have this metadata, so they are excluded to call it.
• removed the anndata's ImplicitModificationWarning which was calling the slow _astype_nansafe() when casting the integer indices to string.

[LucaMarconato]

I'll merge this and then also #373. I spotted a few more places where we could improve performance, I may implement them in a follow up PR.