v0.14.27-dev

Per-platform fleet upgrade

Fixes the fault that stranded every non-darwin host at the last promoted
release while rollouts reported success.

The rollout envelope carries ONE platform (the canary's). The cloud side
corrects that per host from its release ledger — but only for a tag it
has a row for, and release.yml never registered -dev builds. With no
row, the fan-out fell back to the single darwin URL, so linux and
windows hosts probe-failed with "exec format error" across five
consecutive releases while the summary reported "accepted" (which only
ever meant "envelope queued").

release.yml now registers -dev builds with prerelease:true. Registering
is not notifying: GET /api/v1/fleet/target still returns promoted
builds only, so the per-OS QA gate is unchanged.

No daemon code changes — workflow-only.

v0.14.26-dev

outpost v0.14.26-dev

v0.14.25-dev

outpost v0.14.25-dev

v0.14.24-dev

outpost v0.14.24-dev

v0.14.23-dev

v0.14.23-dev

Persist DKS local-path workload storage across runtime recreation and harden kubeconfig refresh against HTML/login/malformed responses.

v0.14.22-dev

outpost v0.14.22-dev

v0.14.21

Promoted from v0.14.21-dev after per-OS QA (windows).
Byte-identical to the tested pre-release — no rebuild.

v0.14.21-dev

outpost v0.14.21-dev

v0.14.20

Promoted from v0.14.20-dev after per-OS QA (windows).
Byte-identical to the tested pre-release — no rebuild.

v0.14.20-dev

v0.14.20-dev

Classify GPU vendors from one shared, tested matcher so an Intel adapter is no longer reported as AMD ("Intel Corporation" contains the "ati" substring).