fix(compress): count tool definitions toward compression trigger

[openjiuwen-sync-bot]

Paired: GitHub #441 ↔ GitCode !2265

Summary

Fixes #​1506: context compression was triggered by comparing message tokens only against tokens_threshold, so a context whose tool definitions already consumed most of the model window could stay over the limit even after compression ran.

Change (openjiuwen/core/context_engine/processor/compressor/current_round_compressor.py)

trigger_add_messages now adds the tool-definition token count (context.get_tools() → token_counter.count_tools(...)) to the running total before comparing against the threshold.

Both access points are guarded so existing behavior is unchanged where the pieces are absent:

• getattr(context, "get_tools", None) — contexts that do not expose get_tools() (e.g. SessionModelContext) skip the tools term entirely (regression fixed after initial attempt crashed them with AttributeError).
• isinstance(tools_tokens, int) — lightweight token counters that only implement count_messages return a non-int via mocks/fallbacks; those are ignored rather than polluting the total.

Tests (tests/unit_tests/core/context_engine/test_current_round_compressor.py)

Two new cases:

• messages below threshold but messages+tools above → trigger_add_messages returns True (regression for #​1506)
• no tools registered → decision unchanged, count_tools never invoked

Verification

• Windows (Python 3.12 + uv): uv run pytest tests/unit_tests/core/context_engine/test_current_round_compressor.py — 10 passed (8 existing + 2 new)
• Linux (Docker, python:3.12-slim + uv): same — 10 passed

Linked Closing Issues:

• Fixes [Bug]: 上下文压缩未计入 system/tool 导致压缩后仍超限 #435

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[openjiuwen-collaboration-bot]

head_sha: b5d5d5115026fcbaeb62fb90e0a333730075b2ea

变更摘要

本次 PR 修复了 #​1506：CurrentRoundCompressor.trigger_add_messages 在判断是否触发压缩时，原先仅统计消息的 token 数，导致工具定义已占用大量上下文窗口的场景下压缩无法及时触发。此变更将工具定义的 token 数也纳入阈值比较，确保 tokens_threshold 能真实反映消息与工具的总开销。

主要改动

• 工具定义 token 计入压缩触发阈值：在 trigger_add_messages 中，通过 context.get_tools() 获取工具列表并调用 token_counter.count_tools() 计算其 token 数，累加到已有的消息 token 总数后再与 _token_threshold 比较。
• 安全访问 get_tools：使用 getattr(context, "get_tools", None) 获取工具访问方法，对于不暴露 get_tools() 的上下文（如 SessionModelContext），静默跳过工具计数，避免 AttributeError。
• 工具计数器返回值类型校验：通过 isinstance(tools_tokens, int) 检查 count_tools 返回值，仅当返回整数时才累加，防止轻量级 token 计数器（仅实现 count_messages）的 mock/fallback 返回值污染总 token 数。
• 新增回归测试用例：test_tools_overhead_counts_toward_trigger_threshold 验证消息低于阈值但消息+工具超过阈值时触发压缩；test_tools_overhead_without_tools_keeps_old_behavior 验证无工具时行为不变且 count_tools 不会被调用。

[openjiuwen-collaboration-bot]

head_sha: b5d5d5115026fcbaeb62fb90e0a333730075b2ea

代码审查

✅ 未发现问题

[openjiuwen-collaboration-bot]

head_sha: b5d5d5115026fcbaeb62fb90e0a333730075b2ea

任务名称	结果	日志操作
静态检查	✅SUCCESS	点此跳转
防投毒检查	✅SUCCESS	点此跳转
开源合规检查	✅SUCCESS	点此跳转
UT测试	✅SUCCESS	点此跳转
ST测试	N/A	N/A
build 编译包	N/A	N/A
ruff codecheck	✅SUCCESS	N/A

[openjiuwen-collaboration-bot]

head_sha: b5d5d5115026fcbaeb62fb90e0a333730075b2ea

任务名称	结果	日志操作
静态检查	✅SUCCESS	点此跳转
防投毒检查	✅SUCCESS	点此跳转
开源合规检查	✅SUCCESS	点此跳转
UT测试	✅SUCCESS	点此跳转
ST测试	N/A	N/A
build 编译包	N/A	N/A
ruff codecheck	✅SUCCESS	N/A