Reset per-report state when reusing a Speedscope renderer

[FanWu-ai]

Reusing a SpeedscopeRenderer carries its event clock and frame-index table into the next report. A second 0.25-second session can emit events from 0.25 to 0.5 while its declared endValue remains 0.25; distinct or empty sessions also retain unrelated frames.

Reset both pieces of per-report state at the start of render. Recursive render_frame calls retain their shared state within that report. Tests compare reused and fresh renderers for identical, different and empty sessions.

Validation on CPython 3.12/Linux:

• Pristine CI split suite executes all 143 tests: 136 ordinary + 7 IPython.
• All 3 regressions fail before the fix; final split suite executes all 146 tests: 139 ordinary + 7 IPython. Complementary skips are the configured isolation between these two modes.
• All 9 pinned hooks pass, including Pyright and the JavaScript bundle rebuild; no generated files changed.
• Docs, native-extension wheel/sdist builds, strict Twine metadata and whitespace checks pass. Docs retain one existing MyST warning; builds retain the existing setuptools license warning.
• 768 reused-render comparisons match fresh instances; independent alternating-session controls pass.

Other interpreter/OS combinations and hosted CI have not run locally.

AI assistance: OpenAI Codex was used to investigate, implement and test this fix. No human review is claimed.

[joerick]

good catch. thank you!