Make the release smoke retry loop actually retry

[jaykomarraju]

0.2.0 published correctly to both registries and the smoke job still went red, for a reason that had nothing to do with the packages.

What happened

PyPI had not propagated tx402==0.2.0 when attempt 1 ran — exactly the race the retry loop exists for. But uv venv had already created .venv before uv pip install failed, so attempts 2 through 5 died at uv venv:

error: Failed to create virtual environment
  Caused by: A virtual environment already exists at: .venv

The && chain short-circuited there, so the install was never retried even once. A loop written to survive propagation lag could not survive its own first failure.

The fix

uv venv --clear recreates the environment each round, so a retry is actually a retry.

Both loops also ran off the end when every attempt failed, leaving the next command to report the problem. That is why the log's headline was Failed to spawn: tx402 — No such file or directory rather than "the package never appeared", which is what actually happened. Each loop now fails explicitly and names the registry and the version.

Verification

Reproduced directly:

	Result
second plain uv venv	already exists — the failure
uv venv --clear	succeeds
uv venv && uv pip install tx402==0.2.0 today	succeeds first try, now that PyPI has propagated

The 0.2.0 artifacts themselves were verified by hand from the registries: all 12 entry points load, both CLIs report 0.2.0, and npm carries a SLSA provenance v1 attestation. tx402/durable-object and tx402/gateway/worker import cloudflare:workers and so load only under the Workers runtime — by design, and confirmed against the published tarball with that specifier stubbed.

CI-only change. No package, source or documentation is touched.