fixed size python string cache

[samuelcolvin]

Use a fully associative cache for strings, rather than a "simple" hashmap, when the number of strings exceeds the hashmap size (currently 500k), this gives a 48% performance improvements according to the 1m_strings python benchmark.

It's currently suffering from this was fixed by @davidhewitt.

thread 'test_python_cache_usage_all' has overflowed its stack
fatal runtime error: stack overflow

in python tests, this can be reproduced locally, and goes away if you remove the GILOnceCell, but that makes things slower (or less secure if you used constant seeds for the hasher).

[codecov]

Codecov Report

Merging #59 (e284786) into main (56db4fc) will decrease coverage by 0.08%.
The diff coverage is 94.56%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #59      +/-   ##
==========================================
- Coverage   96.35%   96.27%   -0.08%     
==========================================
  Files           8        9       +1     
  Lines        1234     1289      +55     
==========================================
+ Hits         1189     1241      +52     
- Misses         45       48       +3     

Files	Coverage Δ
src/python.rs	96.52% <100.00%> (+0.59%)	⬆️
src/py_string_cache.rs	93.97% <93.97%> (ø)

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 56db4fc...e284786. Read the comment docs.

[codspeed-hq]

CodSpeed Performance Report

Merging #59 will not alter performance

_{Comparing fixed-size-cache (e284786) with main (56db4fc)}

Summary

✅ 56 untouched benchmarks

[davidhewitt]

Pushed a tiny commit to use RandomState instead of BuildHasherDefault, this seems to be slightly faster and I think more secure anyway.

[davidhewitt]

All looks great except for one concern about stack usage. It's a shame I can't see a way to allocate a boxed array on stable.

[davidhewitt]

As discussed offline, have a worry here that this line can easily trigger stack overflows. Think we want to avoid accidentally causing import pydantic to cause a stack overflow on some platform with less stack space.

Initial solution sounds like we need to reduce capacity and see how that goes.