feat(sdk): session_reset() for mid-conversation segment boundaries

[OmkarRayAI]

Summary

Ships wikitrace.session_reset() to close one Twitter-feedback gap: when a user clears chat history or a planner restarts mid-trace, callers had no way to mark the boundary. Spans before and after the reset all showed up as one long thread.

API

with wikitrace.session(id="conv-1", user="alice"):
    chain.invoke({"input": q1})              # session_segment=0
    wikitrace.session_reset()                # returns 1
    chain.invoke({"input": "start over"})    # session_segment=1

• Same session_id throughout — cost rollups and user attribution stay grouped
• Distinct session_segment integers so the dashboard can render separate threads
• Outside an active session it's a no-op returning 0 — safe to call from library code

Test plan

• test_session_reset_segments_under_same_session_id — three turns, two resets, assert session_id stable and session_segment increments
• test_session_reset_outside_session_is_noop — returns 0, does not raise
• Full suite: pytest -q tests/ → 98 passed, 14 skipped (no regressions)

What this is NOT

• Not the dashboard tree view (Twitter Q1's other half) — that's a separate, larger PR (~1 day of dashboard work).
• Not subagent cost rollup (Twitter Q2) — also separate.

This is the smallest possible SDK move that lets you reply "shipped" to the segment-boundary part of the feedback today.

🤖 Generated with Claude Code