fix(image): recognise buildx's real "builder exists" error when creating flytex concurrently

[ittus]

Why

flyte deploy builds missing images concurrently. On a host with no flytex builder yet (a fresh CI runner, a new laptop), every build runs docker buildx ls, sees nothing, and runs docker buildx create --name flytex. The second create fails and the whole deploy aborts:

ImageBuildError: Failed to create docker buildx builder 'flytex':
ERROR: existing instance for "flytex" but no append mode, specify the node name to
make changes for existing instances. Try removing it with `docker buildx rm flytex`, ...

#1167 added a guard for this race, but it matches "already exists" in stderr. buildx never prints that for a duplicate create --name. The message comes from builder/builder.go#L485:

errors.Errorf("existing instance for %q but no append mode, specify the node name to make changes for existing instances", name)

already exists is only printed for a duplicate node name inside a builder or a name that clashes with a Docker context. The test added in #1167 fakes the stderr existing instance for "flytex" already exists, which buildx never prints, so the test passes without exercising the real message.

What

• _ensure_buildx_builder reuses the builder when create fails with but no append mode as well as already exists.
• The match is but no append mode rather than the existing instance for prefix. builder.go#L488 also starts with that prefix (existing instance for %q but has mismatched driver %q), and that builder isn't one we should reuse, so it still raises ImageBuildError.
• Nothing changes when create succeeds or fails for another reason.

Test

• Unit test