fix: optimize web generation performance

[DavidJBianco]

Summary

• fix recent connection tuple cache pruning with event-time lazy heap eviction
• add output-equivalence helpers/tests and tuple-cache regression coverage
• update generation performance guidance and worklog
• bump version to 1.8.1 for the dev-to-main release

Validation

• uv run pytest --no-cov --include-slow — 4606 passed, 28 skipped
• uv run pytest --no-cov tests/unit/test_external_parser_docs.py::test_sof_elk_trademark_notice_and_first_references — 1 passed
• focused unit/regression suite — 548 passed
• uv run ruff check .
• uv run ruff format --check .
• 4h before/after SOF-ELK output equivalence — byte-identical
• fixed 12h scenario completed in 44.36s
• fixed full 48h scenario completed in 117.49s

Notes

• eforge eval on the fixed 48h output scored 88/100; the remaining acceptance failures match existing scenario/evaluator behavior documented in the worklog, not the tuple-cache fix.