PERF: get_requirement() raise cache from 2048 to 8192 elements

[morotti]

Hello,

To summarize, the cache was too small :D
Should we have a NEWS item for this? It's very internal with no relevance to end users. EDIT: I added the news item since the CI was red
2% faster pip install on dry run, less in actual run.

You can debug the cache with print(f"parsing {req_string} {get_requirement.cache_info()}")

pip install --dry-run packageswith300dependencies
run with a default size
parsing python-jose[cryptography] CacheInfo(hits=51941, misses=13580, maxsize=2048, currsize=2048)

run with a larger size. 
parsing cryptography>=3.4.0; extra == "crypto" CacheInfo(hits=58758, misses=6947, maxsize=10000, currsize=6692)

before:

after:

Regards.

[notatallshaw]

FWIW, this, I think, was clear when it was changed to 2048: #12663 (comment)

Originally when packaging 22.0+ was going to be vendored the plan was to drop the cache altogether. But as I showed in that PR, there are real world highly esoteric backtracking scenarios where it makes a noticable difference.

I don't have an opinion on if this is a good change or not, but I do think it would be worth running a memory profiler, like memray to see it changes the peak memory usage (though due to #12834 it will be a small fraction of the total).