fix(provider-webdriver): an orientation refused by every route reports dispatched no

[thymikee]

Summary

setWebDriverOrientation tries two routes and, when the driver refuses both as not implemented, threw a COMMAND_FAILED with attempts only: no reason, no dispatched. The router then stamped that failure dispatched: unknown although nothing reached the device. Found by cubic on #3090.

• webdriver-transport.ts: the one owner of the refusal details, unsupportedRouteRefusal() (reason: webdriver_route_unsupported, dispatched: no), used by webdriverError and now by the orientation fallback's final error.
• webdriver-orientation.ts: the every-route-refused error carries those details beside rotation and attempts.
• Golden table: one row, webdriver.orientation.every-route-unsupported → no, driven in webdriver-orientation.test.ts (the owner map gains the webdriver.orientation. prefix). Mutation: dropping the spread fails the row and the existing both-endpoints test.

Part of #3069. +1 row (80).

Validation

webdriver-orientation.test.ts, webdriver-transport.test.ts, contracts/dispatch-disclosure.test.ts (40 tests), pnpm typecheck, check:layering, check:fallow --base 8c0edab4c7, oxfmt on the five files.

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	4.91 MB	4.91 MB	+44 B
Package (unpacked)	4.91 MB	4.91 MB	+44 B
Package (download)	1.47 MB	1.47 MB	-3 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	26.4 ms	25.9 ms	-0.5 ms
CLI --help	78.6 ms	78.9 ms	+0.4 ms

[cubic-dev-ai]

No issues found across 5 files

_{Re-trigger cubic}

[thymikee]

I found no code problems at f4f2ca5. When every orientation route refuses the request, the WebDriver provider now reports dispatched no, which matches the fact that nothing reached the device.

CI is still running: Smoke Tests is in progress and has no failure so far. I did not run anything locally, so the validation list in the PR body is unverified by me. I also did not trace the request router to confirm it keeps a producer-set dispatched no instead of replacing it with unknown. The PR body and the unclassified-dispatch handling in the kernel errors module both point that way.

I did not do a hosted WebDriver run, and I do not think one is needed, because the change only adds error details on a path that never reached the device. No conflicts. Once Smoke Tests finishes green, this is ready, and I am not asking for any code change.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-10-01 11:10 UTC