docs(benchmark): document LongMemEval context_answer/answer_judge steps

[RerankerGuo]

Summary

• Adds end-user docs for the two LME benchmark steps that landed in feat(benchmark): add lme benchmark steps #326 (context_answer_step, answer_judge_step).
• New pages: docs/en/lme_benchmark.md and docs/zh/lme_benchmark.md; both _toc.yml get a new 'Benchmark' section that lists lme_benchmark.
• Documents the per-step input/output contract, the question_type to judge prompt mapping, the answer_judgement normalization rules, and a self-contained job stub plus reme app invocation.
• Calls out that the shipped jinli_lme.yaml does not currently wire the steps into a runnable job, and shows the YAML stanza a benchmark author would add on top.

Why docs only

The runtime code and example config were already merged in #326. grep -r context_answer/answer_judge on docs/ returns nothing, so this PR closes that gap without changing code or example YAML - keeps it orthogonal to whatever follow-ups are planned for #326 / #298.

Verification

• pre-commit run --files docs/en/lme_benchmark.md docs/zh/lme_benchmark.md docs/en/_toc.yml docs/zh/_toc.yml: all hooks green.
• Both _toc.yml files load as valid YAML and reference lme_benchmark.

[RerankerGuo]

Pushed a small follow-up commit a0cb046 on top of the existing branch:

• docs/en/index.md and docs/zh/index.md — adds an 'LME Benchmark' card to the 'Start Reading' grid so the new page is discoverable from the docs landing without having to expand the 'Benchmark' toc section by hand.
• docs/zh/quick_start.md — adds a one-liner right after the auto-memory/auto-resource pointer so users who follow the quick start end-to-end see the benchmark entry point as a follow-up option.

No source code or YAML touched; pre-commit run --files docs/en/index.md docs/zh/index.md docs/zh/quick_start.md stays green.