fix: http action error attribution by splitting request deadlines

[justinkaseman]

Endpoint timeouts were reported as platform execution errors, firing
"[HTTP Action] Execution Errors in More than F nodes" on a user error.

Two causes:

1. The node derived a single deadline from input.Timeout and started it
   before marshal, DON resolution, awaitConnection and SendToGateway.
   The gateway derives the same timeout starting on receipt, so the node
   always expired first. The gateway's correctly classified response
   arrived after responses.cleanup() had run and was dropped, so
   IncrementExternalEndpointError never fired.

2. The ctx.Done() branch returned NewUserError but called
   IncrementExecutionTimeout, which also bumped executionError (Gabriel's fix fix: Execution Errors in More than F nodes not alerting on user errors #755). A
   timeout had nowhere else to land.

Changes:

• Split the deadline. sendCtx keeps input.Timeout and bounds only
  delivery to a gateway. waitCtx is created after the send with
  input.Timeout + ResponseGraceMs (default 5s, mirroring the gateway's
  own send-response budget), so the gateway always reaches its timeout
  first and its classified response wins.

• Split the ctx.Done() branch three ways: TimeoutError (gateway silent,
  system/DeadlineExceeded, counts as an execution error because it now
  genuinely is one), CanceledError (caller cancelled, system/Canceled,
  new http_action_request_canceled_count, not an execution error), and
  the existing classified-response paths, unchanged.

[github-actions]

👋 justinkaseman, thanks for creating this pull request!

To help reviewers, please consider creating future PRs as drafts first. This allows you to self-review and make any final changes before notifying the team.

Once you're ready, you can mark it as "Ready for review" to request feedback. Thanks!

[cl-sonarqube-production]

Quality Gate passed

Issues
1 New issue
0 Fixed issues
0 Accepted issues

Measures
0 Security Hotspots
98.1% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube

[bolekk]

LGTM. I was under the impression that @MStreet3 fixed this problem in #552 but it looks like that PR didn't correctly distinguish between user and system errors and it also didn't split the timeouts between the HTTP request and Gateway not responding at all. @MStreet3 please review this one and correct me if I misunderstood something please.