Add testo/bench benchmarks and document performance trade-offs

[rasuvaeff]

Summary

• Add testo/bench ^0.1.5 to require-dev with testo.php config and benchmarks/ directory
• Two benchmark classes comparing builder API vs direct CompositeSpecification composition
• Document performance characteristics in README.md and llms.txt

Benchmark results (PHP 8.5, Docker, no OPcache)

SpecificationBuilderBench — 7-step chain (5 conditions + limit + offset):

	Mean*	RStDev*
SpecificationBuilder (current)	3.25µs	±0.58%
CompositeSpecification direct	2.41µs (−26%)	±1.57%

Overhead = 7 clone $this operations (one per builder step).

OrCompositionBench — OR with 3 branches:

	Mean*	RStDev*
orWhere() callback	2.78µs	±0.45%
OrSpecification::create() direct	1.43µs (−48%)	±0.80%

Overhead = temporary mutable builder + closure invocation per orWhere() call.

All RStDev < 2% — results are stable.

Why this matters

For typical web requests (1–5 specs, DB query ≫ builder cost) the overhead is negligible.
The documentation clarifies when to prefer the direct API (high-throughput batch loops).

Run benchmarks

composer bench
# or
make bench