A gateway whose dependency environment lacks aiohttp (a sealed env from
`hermes update` with a partial messaging extra, #126358) spawned the bridge,
then every /health poll raised ModuleNotFoundError inside
`_poll_bridge_health`'s blanket except. connect() returned False with no
fatal code, so the reconnect watcher retried forever while the bridge
answered /health 200, and telemetry counted each attempt as error_class
"other".
- connect() now obtains aiohttp through PM (`ensure_import("sms")`, the
extra that is exactly aiohttp) before starting the bridge, the same
lazy-install path Slack/Teams use. When PM cannot install it (lazy
installs disabled, foreign interpreter, install failure) or aiohttp is a
half-removed namespace shell (#71308), it fails with the named,
non-retryable `whatsapp_aiohttp_missing` and never spawns the bridge.
- The bridge-died / "did not start in 15s" paths set a retryable fatal code
(`whatsapp_bridge_exited` / `whatsapp_bridge_timeout`), so reconnect
status and connect telemetry name the cause instead of "other".
- Folds the salvaged test file into test_whatsapp_connect.py as two
invariant tests.
2dda207409