fix: improve uv run completions

[TATP-233]

Summary

• Include project scripts and runnable benchmark/script paths in uv run completion metadata.
• Defer to shell file completion when completion metadata has no candidates for non-training commands.
• Add completion tests for project scripts, hierarchical run paths, and unknown command arguments.
• Adjust benchmark plot y-axes to log scale and narrow backend benchmark default sizes.

Validation

• uv run pytest tests/test_completion.py
• make test-all