Speedup of peptide parsing and annotation

[jspaezp]

This PR implements 4 main things, all with the purpose of improving speed of spectrum annotation workflows.

1. A fast-pass for unmodified peptides during the parsing.
2. The option for a simpler parsing grammar.
3. LRU caching of the parser (read once per session, not once per parse of a proforma sequence)
4. The option to annotate spectra passing a list of proteoforms directly (instead of a sequence)
   • This feature is critical for me, since I have a workflow that uses both the proteoforms directly and the annotated spectra. Therefore by itself makes my workflow 2x faster.

Benchmarks

Using some dummy peptide examples the speedup i see in the parsing is:

With mods

29.51it/s -> (baseline), greedy loading, no fastpass
137.54it/s -> + unmod fastpass, cached full parser (4x improve)
168.48it/s -> + simple parser (1.22x improve,~6x from baseline)

Without mods

34.18it/s -> (baseline) greedy loading, no fastpass
995089.92it/s -> + unmod fastpass, cached full parser (~ 30000x improve)
1081006.19it/s -> + simple parser (equivalent for practical purposes)

On a heavy annotation workflow I have these changes dropped the run time from 45 mins to 2.20 :P

LMK what you think!
Best

[jspaezp]

btw the tests that involve reading from USI are also breaking on master on my local system.

[bittremieux]

Thanks for the contribution. The fast-pass for unmodified peptides is a great addition, as well as annotating proteoforms directly. I have a few comments related to the code that should be relatively easy fixes.

I'm a bit less convinced about the simplified grammar though. This can be discussed in a bit more detail, see the respective comments.

[jspaezp]

@bittremieux added the suggestions, LMK what you think!