fix(generate): skip local model validation when default_remote is set

[aladac]

Symptom

$ tsr generate -m obsessiveCompulsive_v20.safetensors "a cat on a table"
Model 'obsessiveCompulsive_v20.safetensors' not available on ComfyUI host
(looked in checkpoints/ — 2 entries)
Did you mean:
  flux_dev.safetensors

…even though config has default_remote = "runpod" and the model exists on the pod.

Root cause

Two validation gates in generate() / _run_generation() used the raw remote CLI flag instead of the resolved remote URL:

line	guard
cli.py:1083	if model and not remote:
cli.py:1403	if model and not json_output and not remote:

When -r is omitted but default_remote is set in config, both treat the run as local and call _validate_model_available → get_loaded_models() against the local [comfyui].url (often a stale LAN address or empty). The actual dispatch ~100 lines later DOES respect default_remote, so the request would have worked if validation hadn't fired first.

Fix

Both gates now use do_resolve_remote(remote) — returns the resolved URL when default_remote is set even if -r was omitted, so the local validation is skipped and the server-side validator on the remote tensors API takes over. Matches the intent already documented in the comment block: "Skipped in --json mode and for remote dispatches (server already validates)."

Also dedupes the do_resolve_remote import in _run_generation — moved to top of function instead of inline mid-body.

Validation

• uv run pytest tests/ -x → 374 passed
• uv run ruff check tensors/cli.py → All checks passed
• uv run ruff format --check tensors/cli.py → already formatted
• uv run mypy tensors/cli.py → no issues
• End-to-end: tsr generate -m obsessiveCompulsive_v20.safetensors "a cat on a table" -o /tmp/cat.png from chi@fuji (no -r flag) → 1024×1024 PNG written, dispatched through pod's tensors API.