AI: Secure imported contexts and hide page-touching tools

Imported Lighthouse reports and Performance traces are untrusted and
could contain prompt injections designed to exploit the AI assistant.
To prevent cross-origin script execution and data exfiltration, this
change disables page-touching tools when operating on imported
contexts.

For Lighthouse:
- Tag imported reports with `isImported = true` upon loading.
- Restrict the Accessibility Agent to only declare `getLighthouseAudits`
  when the report is imported, disabling all page-touching tools.
- Prepend a security warning to the LLM query for imported reports.

For Traces:
- Dynamically hide the `getFunctionCode` tool in the Performance Agent
  if the trace is not fresh, preventing any execution attempts.
- Prepend a security warning to the LLM query for non-fresh traces.
- Prevent the security warning and the user query from leaking into the
  user-facing "Used Context" UI disclosure.
- Delete the dead `external` flag from `PerformanceTraceContext` and
  associated guards, enabling full linkification for active traces.

Fixed: 513711812
Change-Id: I22d26e6ac1fd26bb0772b3eb4f9165f548532bc1
Reviewed-on: https://chromium-review.googlesource.com/c/devtools/devtools-frontend/+/7902589
Auto-Submit: Jack Franklin <jacktfranklin@chromium.org>
Reviewed-by: Paul Irish <paulirish@chromium.org>
Commit-Queue: Jack Franklin <jacktfranklin@chromium.org>