evidence: witness kimi_code on a real run

[jothimani-rajendran]

kimi_code's matrix row currently rests on vendor-docs (recorded against CLI), not on
anyone watching the agent run. That is an honest label, and it is also the gap this project
exists to close: a capability nobody re-ran is a claim, not a measurement.

If you have kimi_code installed, you can close it in about two minutes. You are better placed
to than a maintainer who does not hold that licence.

python3 tools/capture.py detect
python3 tools/capture.py install --agent kimi_code
# ...use kimi_code normally for a minute...
python3 tools/capture.py report
python3 tools/capture.py uninstall --agent kimi_code

Then open an issue with the Evidence report template and paste the result.

Things worth knowing before you start, all covered in CONTRIBUTING.md:

• The capture probe always allows. It records and gets out of the way, so it cannot break
  a session.
• Payloads are reduced to shape before anything touches disk -- keys and types survive,
  values become markers like <str:41>.
• A result that contradicts the matrix is the most valuable thing you can send. We are not
  collecting confirmations. If your run says kimi_code behaves differently from what we publish,
  that is a row we are getting wrong in public.
• A hook that never fires is still a finding -- it means the config path or format is wrong
  for your version, which is exactly what documentation does not tell you. Report it with the
  version.

No code required. Report the agent's version alongside the result.