Maintain benchmark suite and publish reproducible comparisons

Current status

The repository now has Go benchmarks for parsing, rendering, drops, expressions, and filters. The stale engine benchmark templates were repaired in #148, and the full suite executes with:

cgo-test -run "^$" -bench . ./...

This issue remains open for turning those benchmarks into a maintained, useful performance suite.

Remaining work

• Select representative fixtures for parsing and rendering rather than relying only on small synthetic templates.
• Add a fast CI smoke check that compiles and executes each benchmark once, without treating noisy timing changes as regressions.
• Document reproducible local benchmark commands, including benchstat comparisons.
• Record Go version, OS, CPU, benchmark count, and dependency versions with published results.
• Evaluate cross-engine comparisons against current Liquid implementations using equivalent fixtures and clearly documented semantics.
• Keep performance claims separate from correctness and compatibility claims.

References

• https://github.com/SlinSo/goTemplateBenchmark
• https://github.com/Shopify/liquid/wiki/Ports-of-Liquid-to-other-environments

Start by running cgo-test -run "^$" -bench . ./... to inspect the existing Go benchmark suite, then review the referenced comparison projects and remaining-work list. Done means representative fixtures, a non-regression CI smoke check, reproducible benchmark documentation, recorded environment metadata, and clearly scoped cross-engine comparisons.