fix(cli): Propagate bundle render output errors

[reneleonhardt]

What

Make a successful timoni bundle build exit guarantee that the complete in-memory render was accepted by stdout. Propagate the final writer error, and report io.ErrShortWrite when the writer accepts fewer bytes without returning an error. The command, flags, rendered bytes, ordering, concurrency, and existing --output-dir behavior stay unchanged; the --output-dir path already returned file write errors and is untouched.

Why

Bundle rendering already has an excellent automation shape: Timoni builds every instance in memory, preserves bundle order, and performs one final stdout write only after the whole render succeeds. The final Write result is currently discarded, so a full, closed, or otherwise failing output sink can receive incomplete data while Timoni reports success. This fix turns the exit status into a dependable completeness signal for artifact pipelines and other render-only consumers, without adding a destination directory, staging tree, or apply step.

Reproduction

The command before and after this change is identical.

# Linux: a failing output sink (writes fail with ENOSPC).
timoni bundle build -f bundle.cue >/dev/full
echo $?

# main: discards the write error, leaves the stdout stream empty, still exits 0.
# this PR: returns the write error, logs it to stderr, exits 1.

What the reviewer must decide

1. Whether failing the command on a short write with a nil error (not only on a returned writer error) is the right completeness contract for a render-only stdout command.
2. Whether keeping the sibling single-instance timoni build on its existing behavior (it already propagates the write error but does not check short writes) is acceptable, or whether the two render commands should share the same check.

Verification

go test ./cmd/timoni/ -run 'Test_BundleBuild' -count=1

The focused suite covers a sentinel writer failure, a nil-error short write, an ordinary render failure with zero output, existing stdout bytes and ordering, and unchanged directory output.

Developed with carefully directed, manually reviewed AI assistance.

Co-Authored-By: GPT-5.6 Sol codex@openai.com

[stefanprodan]

LGTM

Thanks @reneleonhardt