Function estimatePostTrimTokens

Estimate post-trim token count from the session analysis breakdown.

Trim rules (mirrors trimmer.ts):

1. Pre-compaction content → skipped entirely
2. file-history-snapshot → removed entirely
3. queue-operation → removed entirely
4. Image blocks in tool results → stripped always
5. tool_result > threshold → stubbed (~35 bytes each)
6. tool_use input for write ops + broad fallback → stubbed
7. thinking blocks → removed entirely
8. API usage metadata → stripped

We estimate that ~70% of tool result bytes come from results > 500 chars (conservative — real sessions often see 85-95%). We estimate ~30% of tool_use request bytes are from stubbable input fields (Write/Edit payloads, Task prompts, etc.).