feat(feishu): add model-aware context guard defaults for long chats

[slicenferqin]

Summary

This PR addresses #279 by adding a practical context-control path for Feishu sessions in ClawX.

• Add optional historyLimit field to Feishu channel settings UI.
• Normalize historyLimit as a non-negative integer when saving config.
• Auto-apply safe defaults for long-running Feishu conversations (only when missing):
  • agents.defaults.compaction.mode = safeguard
  • model-aware reserveTokensFloor (≈5% of context window, with sane bounds)
  • memoryFlush.enabled = true + model-aware softThresholdTokens
  • contextPruning defaults (cache-ttl, softTrim, hardClear)
• Keep manual overrides intact (existing user config is not overwritten).
• Add i18n labels/descriptions for the new Feishu setting in zh/en/ja.

Why

Issue #279 reports token explosions (30k~50k) in Feishu-integrated flows that can crash local models. This change adds both:

1. an explicit user option (historyLimit), and
2. a no-touch auto-protection baseline for users who never tune advanced settings.

Validation

• pnpm install
• pnpm -s typecheck

Fixes #279