LFQ2: Faster and much better

[lazear]

This PR completely rewrites Sage's existing (and relatively amateurish) label free quantification code.

Highlights of the new algorithm:

• Hybrid (global regression alignment + local feature-specific correlation optimized offsets) chromatographic alignment
• Directed ion current extraction across charge states and isotopologues within m/z and RT window
• Calculation of normalized spectral angle of observed isotope distribution vs expected
• Local maxima detection with peak merging/splitting based on the union-find algorithm
• Extremely fast - the above steps typically take < 30s for the 20 file IonStar dataset

Benchmarks are being performed on the following datasets:

• PXD001819 (UPS1 spike in to yeast background)
• PXD003881 (IonStar E.coli/human proteome mixture)
• PXD028735 (LFQ Bench 2022 E.coli/human/yeast proteome mixtures)

Current benchmarks

Remaining To-Do Items:

• [x] Compare performance of Sage's LFQ against MaxQuant, FlashLFQ, or perhaps something else?
• [ ] Write some test cases of peak picking, integraton
• [x] Investigate whether the current algorithm is compatible with some kind of decoy approach
• [x] Find the proper trade off between missing values/sensitivity/accuracy
• [x] Rebase & clean up commit history

[lazear]

Current algorithm can be modified to perform target/decoy approach for peak integration, a la IonQuant (make a decoy PrecursorRange entry that is mass-offset).

Peak score & spectral angle appear well-calibrated and sufficient to use for TDS, if desired. (apex RT in grid-space is also a great feature)

label = decoy

[lazear]

Time to merge! Would be good to still write some test cases for automated testing, but I have done enough manual testing of the peak integration to satisfy myself for now 😄