[kernel-1116] browser events: add external events

[archandatta]

As per plan -> #217

Note

Medium Risk
Adds new public API endpoints for injecting and streaming events and changes capture-session stop behavior to emit terminal frames; issues here could impact event delivery semantics and client reconnect behavior.

Overview
Adds a new external events API: POST /events/publish to inject validated events into the active capture session (with reserved types blocked and source.kind forced to kernel_api), and GET /events/stream to stream capture envelopes over SSE with Last-Event-ID resume, keepalives, and explicit handling for dropped events.

Updates the events pipeline to support this streaming lifecycle by introducing CaptureSession.Active, publishing a synthetic session_ended envelope on Stop, and adding an unfiltered publish path (PublishUnfiltered) for externally supplied events. OpenAPI/oapi-codegen output is regenerated accordingly (new schemas/client methods, SSE headers), and tests are updated for renamed enum constants.

^{Reviewed by Cursor Bugbot for commit f97e3a9. Bugbot is set up for automated code reviews on this repo. Configure here.}

[firetiger-agent]

Firetiger deploy monitoring skipped

This PR didn't match the auto-monitor filter configured on your GitHub connection:

Any PR that changes the kernel API. Monitor changes to API endpoints (packages/api/cmd/api/) and Temporal workflows (packages/api/lib/temporal) in the kernel repo

Reason: PR title and branch suggest browser events feature work, but no file changes are provided to confirm modifications to API endpoints or Temporal workflows.

To monitor this PR anyway, reply with @firetiger monitor this.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 0a89c58. Configure here.}

[Sayan-]

matches plan well and deviations from it are reasonable!

[Sayan-]

should seq be monotonic across start/stop cycles? If so, I think Start shouldn't reset it here. That would make seq represent the process-wide event stream position rather than the position within a single capture session.

[Sayan-]

could we add one lifecycle test for the new event flow? Something like start capture session, publish an event through PublishEvent, read it from StreamEvents, then stop and verify the stream receives session_ended.