Fix nil derefs in job run logging and tmpdir leak in terraform import

[arsenyinfo]

Changes

Three small bug fixes surfaced by a code review pass:

• bundle/run/job.go — guard task.State in isFailed/isSuccess and run.State in logFailedTasks against nil. The same file's onProgress already does this, so the inconsistency is a latent panic if the Jobs API ever returns a run/task with no state.
• bundle/run/job.go — in the --no-wait path, check the error from GetRun before dereferencing details.RunPageUrl. Previously a failed GetRun would panic on the next line.
• bundle/deploy/terraform/import.go — move defer os.RemoveAll(tmpDir) to immediately after MkdirTemp succeeds. It was placed after two early-return paths (tf.Import / tf.Plan failures), so the temp dir leaked on those failures.

Why

All three are unambiguous defects: two nil-pointer panic sites and one resource leak. No behavior change on the happy path.

Tests

• Existing tests in bundle/run and bundle/deploy/terraform cover the happy paths and continue to pass behaviorally; the defended branches were previously panic / leak paths with no coverage.
• Local go build is currently blocked by a flaky toolchain proxy (503 on proxy.golang.org); relying on CI to validate.

[pietern]

Thanks!

This doesn't seem to be user facing, so I removed the changelog entry.