Attach MCP shutdown handlers before async startup, add broken-pipe and
crash handling, and explicitly stop the Xcode watcher during teardown.
This closes lifecycle gaps that could leave MCP processes running after
clients disappear and adds concrete shutdown reasons for investigation.

Add lifecycle Sentry metrics, richer session diagnostics, and a repro
script so process age, peer counts, memory, and active runtime state are
visible when this issue recurs in the wild.

Fixes #273
Co-Authored-By: Codex <noreply@openai.com>

When the MCP parent process exits abruptly with in-flight requests, shutdown can
block indefinitely waiting for server.close(). This keeps npm exec wrapper
processes alive and allows the child node process to climb in RSS.

Add a transport-disconnect fast-exit path that detaches lifecycle handlers,
races server.close() against a short timeout, and then exits immediately.
This preserves normal shutdown behavior for other reasons and keeps Sentry
enabled.

Also add focused unit tests for the bounded close helper outcomes.

Fixes #273

Update @modelcontextprotocol/sdk from ^1.25.1 to ^1.27.1 so local
and harness testing run on the latest SDK behavior.

Add /repros to .gitignore to keep lifecycle repro artifacts out of
source control during repeated leak investigations.

Refs #273
Co-Authored-By: Codex <noreply@openai.com>

Move MCP shutdown orchestration into a dedicated bounded runner and stop
using Sentry.close in MCP shutdown paths.

Suppress stdio immediately on transport disconnect, capture one shutdown
summary event, seal further capture, then perform bounded Sentry.flush
before explicit exit. Add shared bulk cleanup helpers for tracked runtime
resources and wire shutdown through those helpers.

This keeps shutdown deterministic under parent/disconnect teardown while
preserving high-value shutdown telemetry.

Co-Authored-By: OpenAI Codex <noreply@openai.com>

…pshot map keys during iteration