[pavog]

There was no way to measure how long parsing or analysing a log takes, so performance changes could only be estimated.

This adds phpbench benchmarks for the PatternParser and PatternAnalyser that run on generated logs.
The logs are generated with a fixed seed, so every run and branch measures the same content without checking large files into the repository.
The analyser benchmark uses up to 20 distinct insight classes because real users register many of them.
The benchmarks are meant to compare a branch against master locally with --tag and --ref.
CI only checks that they still run, because timings on shared runners vary too much for a threshold.

Changes

• add phpbench as dev dependency with a composer bench script
• add a deterministic log generator for benchmark input
• add benchmarks for parsing and analysing
• add keyword insight classes to analyse with many distinct insights
• run the benchmarks once in ci as a smoke test
• document how to compare benchmarks against master