feat(supervisor): record launch and exit of sandbox sessions - #5252

Summary

The entrypoint's launch reached only the sandbox container's stdout, and no process the runtime started had an exit event. The runtime now keeps a bounded log of the launches and exits of the entrypoint and each exec or SSH session, and the supervisor emits them as OCSF Process Activity.

Second of a stack of four for #4268 (based on #5251):

1. feat(supervisor): identify the process behind each network event #5251 Identify the process behind each network event
2. feat(supervisor): record launch and exit of sandbox sessions #5252 This PR: record launch and exit of sandbox sessions
3. feat(supervisor): record each exec in the workload and its exit #5253 Record each exec in the workload and its exit
4. feat(supervisor): record the command line of each process #5254 Record the command line of each process

Related Issue

Refs #4268

Changes

• The runtime records the launch and exit of every process it spawns and reaps (entrypoint, exec and SSH sessions) in a bounded, sequence-numbered log.
• New ReadProcessEvents long-poll RPC on the sandbox protocol (capability openshell.sandbox.process-events): sequence cursor, size-capped responses, and a count of events no longer retained.
• The supervisor reads it for the life of the boundary and emits Process Activity Launch (pid, start time, parent) and Terminate (exit code; 128 + signal for signal exits).
• Each event keeps the runtime's observation time and a deterministic metadata.uid derived from the process instance, so a replacement supervisor's re-read produces duplicates that consumers can drop.

Before: no Process Activity in the supervisor's OCSF output.

After:

PROC:LAUNCH    [INFO] /usr/bin/bash(78)
NET:OPEN       [INFO] ALLOWED /usr/bin/bash(82) -> example.com:443 [policy:demo engine:opa]
PROC:TERMINATE [INFO] /usr/bin/bash(78) [exit:0]
PROC:LAUNCH    [INFO] /usr/bin/bash(91)
PROC:TERMINATE [INFO] /usr/bin/bash(91) [exit:3]
PROC:LAUNCH    [INFO] /usr/bin/bash(93)
PROC:TERMINATE [INFO] /usr/bin/bash(93) [exit:137]

{"activity_name":"Terminate","message":"Process exited: pid=93 signal=9","exit_code":137,"process":{"name":"/usr/bin/bash","pid":93,"created_time":1793758508670,"parent_pid":1},"actor":{"name":"openshell-sandbox","pid":1}}

The network actor (pid 82, parent 78) joins to the session's Launch on pid + created_time.

Known limitations:

• Terminate reports status: Success for any observed exit, including non-zero and signal exits, so filtering on status=Failure won't find crashed sessions. Use exit_code.
• The runtime still writes its own entrypoint Launch to container stdout for older supervisors, so stdout collectors see two entrypoint Launches.
• After a supervisor takeover the log is re-read; duplicates repeat metadata.uid and the original time.
• There is no bounded drain at teardown; the final exit relies on the long poll waking first, which it does in practice.
• created_time is within about 1 s of wall-clock time (btime has whole-second resolution). It is consistent across process and network events, so it still works as a join key.
• A lost > 0 count is only a warning log here; PR 3 turns it into a record.

Testing

• Lockfile checks, rust:format:check, and every CI clippy invocation (--locked, including perf-harness) pass on macOS and the aarch64 Linux target for this commit
• Unit tests added/updated; sandbox, sandbox-backend, and supervisor suites pass in an Ubuntu 24.04 guest (kernel 6.8)
• Manual end-to-end: gateway + Docker sandbox (kernel 7.0), before/after OCSF output above

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)
• Architecture docs updated (if applicable)