fix(node): harden websocket transport reliability

[h3zh1]

Summary

• reset exponential reconnect backoff only after an accepted WebSocket session remains stable
• detect silent peers with bounded ping, pong, read, and write deadlines
• report actionable TLS, DNS, TCP, HTTP handshake, timeout, and WebSocket close diagnostics
• advertise a process-scoped node identity that stays stable across reconnects and changes after restart
• preserve correlated enrollment rejection codes and messages returned by the server

User-visible behavior

• repeated short-lived connections continue backing off instead of resetting after every handshake
• half-open WebSocket connections are detected and reconnected without waiting indefinitely
• certificate, token, endpoint, network, and peer-close failures produce specific Runner logs
• AOP hubs can distinguish a reconnecting process from a different process reusing the same node ID
• enrollment conflicts such as ALREADY_EXISTS reach the Runner log instead of becoming a generic acceptance error

Commit structure

1. fix(node): reset reconnect backoff after stable sessions
2. fix(node): enforce websocket liveness deadlines
3. fix(node): classify websocket connection failures
4. feat(node): advertise runner process instance identity
5. fix(node): surface enrollment rejection reasons

Verification

• go test ./pkg/node ./core/deps at each commit
• go test -race -count=1 ./pkg/node
• go vet ./...
• go mod tidy -diff
• git diff --check origin/master...HEAD

Test note

The unfiltered go test ./... run on macOS still fails the pre-existing TestShellCommandRedirectionAndInvocationContext assertion because the shell resolves /var/... as /private/var/.... The same failure reproduces on unmodified origin/master; all pkg/node tests pass.

Integration

Cairn must update its AIScan module version after this branch is published or merged before it consumes the final commit hashes.

Deferred

• the previously recorded WebSocket 512 KiB message limit is not addressed here
• Cairn Server connection-pool and enrollment-policy changes remain in the Cairn repository

[M09Ic]

Pushed bd6e7c6d to simplify the runner WebSocket reliability changes after auditing every new package variable, connection field, config field, and function parameter.

Key changes:

• removed the mutable reconnect timing hooks and the test-only Dialer / Liveness fields from production config
• reduced webSocketEnvelopeStream lifecycle state to the connection, write mutex, and codec mode; heartbeat lifetime is now owned by the connection context
• close the stream after an application write failure so a blocked Recv wakes immediately and reconnects, instead of waiting for the read deadline
• calculate ping deadlines from time.Now() at the actual write rather than from a possibly stale ticker timestamp
• generate instance_id inside each RunToolNode invocation, keeping it stable across that runner's reconnects without sharing mutable process-global identity
• keep fixed liveness/backoff values as constants and test behavior without production timing injection
• preserve raw DNS/TCP/timeout errors while retaining actionable TLS, HTTP handshake, WebSocket close, and enrollment rejection diagnostics

The cleanup removes 220 net lines from the previous PR head. The PR is now 7 changed files with +599/-19, down from 9 files with +818/-18.

Verified locally:

• go test -race -count=1 ./pkg/node
• go test -count=1 ./core/deps
• go vet ./...
• go mod tidy -diff
• git diff --check

[M09Ic]

CI follow-up: the new workflow run is blocked during actions/checkout, before any AIScan lint/test command runs. The current master merge result references web/frontend/cyber-ui commit 2a00bac673570fef65535c4753ffc4f6a95ff502, which the submodule remote reports as "not our ref".

This reproduces on the current base commit 0b4850a6: its own CI run failed at the same checkout step: https://github.com/chainreactors/aiscan/actions/runs/31357987246

PR run: https://github.com/chainreactors/aiscan/actions/runs/31359589270

The runner changes themselves continue to pass the local race, tests, vet, tidy, and diff checks listed above. I have not changed the unrelated submodule pointer in this PR.

[M09Ic]

Follow-up cleanup pushed in e2d5c26d, focused specifically on recovery from a blocked transport rather than adding a tunable liveness framework.

• only two independent policy values remain: a 30-second keepalive interval and a 10-second write-block timeout
• websocketPongWait is derived as three heartbeat intervals (90 seconds)
• reconnectStableAfter is derived as one complete liveness window plus one heartbeat interval (120 seconds), so a fully silent peer times out before it can reset reconnect backoff
• removed the stream write mutex: application envelopes already pass through one sendCh writer, matching the EnvelopeStream single-Send contract; Gorilla explicitly permits WriteControl concurrently
• the recovery path remains singular: a blocked write reaches its deadline, the write error closes the stream, blocked Recv wakes, and the outer loop reconnects with backoff

Re-verified with go test -race -count=1 ./pkg/node, go test -count=1 ./core/deps, go vet ./..., go mod tidy -diff, and git diff --check.