perf: optimise <v> cell value parsing

[alexander-beedie]

Builds on #611 (optimising xlsx cell value reads) for a further ~8-10% performance gain for typical reads.

Optimisation

When reading "n", "s", "b", and "e" values with read_value, these types are guaranteed to be plain ASCII so we can bypass the xml10_content and UTF8/String overhead and deal directly with them as raw bytes. (Other types fall-through to the usual path, with the necessary additional handling).

• s: integer index (into shared string table)
• n: numeric literal
• b: boolean
• e: error

Performance

Note: these tests are on mixed-type data; if you have a workbook with nothing but strings that aren't in the shared string table, you won't see the same gains. These timings should be fairly representative of real world data though.

The gain here is additive with #621 (the code being updated is entirely disjoint), so when combined we're looking at about another ~25% speedup¹ over the latest release for typical workbooks.

┌──────────────────┬──────────────┬─────────────┬─────────────┬─────────┐
│ file             │        cells │ master (ms) │ branch (ms) │ speedup │
├──────────────────┼──────────────┼─────────────┼─────────────┼─────────┤
│ test_1000000_100 │  100,000,000 │      22,055 │      19,913 │    9.7% │
├──────────────────┼──────────────┼─────────────┼─────────────┼─────────┤
│ test_100000_100  │   10,000,000 │       2,200 │       1,983 │    9.9% │
├──────────────────┼──────────────┼─────────────┼─────────────┼─────────┤
│ test_10000_100   │    1,000,000 │       215.6 │       197.7 │    8.3% │
├──────────────────┼──────────────┼─────────────┼─────────────┼─────────┤
│ test_10000_10    │      100,000 │        22.4 │        20.3 │    9.5% │
├──────────────────┼──────────────┼─────────────┼─────────────┼─────────┤
│ test_1000_10     │       10,000 │        2.30 │        2.08 │    9.7% │
└──────────────────┴──────────────┴─────────────┴─────────────┴─────────┘

Footnotes

1. Benchmarked on: Apple Silicon M3 Max ↩