Flush managed runs when child processes exit

[Stephen Belanger (Qard)]

Summary

• tag events from each managed child process tree with an invocation-local ID
• group accepted daemon sessions by invocation and flush them after the child exits
• propagate the grouping through command hooks and shared JavaScript adapters without flushing unrelated sessions

Why

bt trace run previously depended on terminal agent hooks to trigger delivery. If the child exited nonzero, was terminated by a signal, or was interrupted before a terminal hook, the managed run could return without an awaited backend flush.

Impact

Managed runs now perform a scoped final flush after every child exit path, including success, nonzero exit, signal termination, and parent interruption. Each session gets a 10-second flush timeout. Flush failures remain fail-open and do not replace the child's exit status.

Validation

• make test
• cargo test --manifest-path bt-daemon/Cargo.toml --all-features --locked
• cargo clippy --manifest-path bt-daemon/Cargo.toml --all-targets --all-features --locked -- -D warnings
• cargo fmt --manifest-path bt-daemon/Cargo.toml -- --check
• pnpm test in src/runtime/js-daemon-client
• git diff --check