Add formula token cache to reduce repeated parsing overhead

[kemo]

Summary

• Adds a static bounded formula token cache (max 1000 entries) to the Calculation engine
• Before parsing a formula, parseFormula() checks the cache — on hit, returns cached tokens immediately, skipping all regex operations
• When the cache reaches capacity, it is fully cleared (simple eviction)
• Provides clearFormulaTokenCache() and getFormulaTokenCacheSize() static methods for testing and memory management

Changes

• src/PhpSpreadsheet/Calculation/Calculation.php: Static $formulaTokenCache array, cache lookup in parseFormula(), clear/size methods
• tests/PhpSpreadsheetTests/Calculation/FormulaTokenCacheTest.php: 8 tests covering correctness, cache reuse, clearing, complex formulas, eviction
• tests/PhpSpreadsheetTests/Benchmark/FormulaTokenCacheBenchmark.php: Benchmark comparing cache hit vs miss for 1K formula cells and 10K parseFormula calls

Test plan

• Verify formula parsing produces identical results with caching
• Verify identical formulas reuse cached tokens
• Verify the cache can be cleared and rebuilt
• Verify cache eviction at 1000 entries
• Run benchmark: vendor/bin/phpunit --group benchmark --filter FormulaTokenCacheBenchmark --stderr

[kemo]

Benchmark Results (formula-ast-parsing)

Targeted benchmark: parse repeated formula tokens (500 distinct patterns).

Benchmark	Master	Branch	Speedup
10k formulas cold	263ms	17ms	15x
10k formulas warm	261ms	1.4ms	186x
50k formulas cold	1,366ms	21ms	65x
50k formulas warm	1,345ms	7ms	192x
100k formulas cold	2,625ms	28ms	94x
100k formulas warm	2,726ms	14ms	195x
Calculate 1k cells	168ms	165ms	~same
Calculate 5k cells	795ms	805ms	~same

Enormous parsing speedup (15–195x). Calculation time unchanged — the win is entirely in token parsing, not evaluation. Memory neutral.

[oleibman]

I strive to avoid the introduction of new static properties. Is it possible to make $formulaTokenCacheMaxSize and $formulaTokenCache instance variables rather than static? And add a routine to set the value of the former? And, arguably, make it default to zero, making its use opt-in ( the results are impressive but I'm nervous)?

[kemo]

Makes sense — I'll refactor both to instance properties on the FormulaParser and default the cache max size to 0 so it's opt-in. Users who want the parsing speedup can explicitly enable it.

[oleibman]

Thank you for your contribution.