Polish CLI UX and speed up local preview flows

[TrevorBasinger]

Summary

This PR is a CLI polish and startup/performance pass for roar.

It focuses on two things:

1. making common CLI behavior more trustworthy and easier to recover from
2. making local read-only and preview flows feel fast by keeping them local and cutting unnecessary startup work

What Changed

CLI UX and correctness

• return non-zero exit codes for user-facing show, log, and status errors instead of printing failure text as success
• fix roar init --path so .gitignore handling uses the target repo, not the caller repo
• fix recovery/help text, including removing the stale roar clean guidance
• add explicit roar show --path, --job, --artifact, and --session selectors
• make lazy subcommand import failures surface as Click errors instead of Python tracebacks
• promote roar auth key as the primary SSH key workflow and keep auth register as a compatibility alias
• improve top-level help and success output for run, put, and register

Performance and profiling framework

• add a reusable profiling harness at scripts/profile_roar.py
• add profiling docs at docs/developer/profiling.md
• lazy-load query, DB, presenter, core model, config, and GLaaS integration surfaces to reduce startup work
• add a sqlite-backed read-only query context for status, log, and show
• keep put --dry-run local by skipping backend/runtime/git setup
• keep register --dry-run local with preview-only runtime, raw config access, lightweight git/session prep, and lightweight job normalization
• add regression coverage around lazy import boundaries and preview-path behavior

Profiling Results

Latest full profile sweep from this branch:

• cli_help: 47.1ms
• cli_run_simple: 985.2ms
• cli_status_active: 70.3ms
• cli_show_session: 82.1ms
• cli_register_dry_run: 111.2ms
• cli_put_dry_run: 77.6ms

Largest wins from the original branch baseline:

• cli_put_dry_run: 3131.6ms -> 77.6ms
• cli_status_active: 491.0ms -> 70.3ms
• cli_show_session: 445.2ms -> 82.1ms
• cli_register_dry_run: 424.5ms -> 111.2ms

The main remaining product-path hotspot is still roar run, which is dominated by tracer/process and provenance collection work rather than generic CLI startup.

Verification

Ran from /home/trevor/dev/roar-cli-polish:

uv run --extra dev ruff check .
uv run --extra dev mypy roar
uv run --extra dev pytest -m "not live_glaas and not ebpf"
uv run --extra dev python scripts/profile_roar.py --scenario cli_register_dry_run --iterations 5 --top 20 --output tests/benchmarks/results/profile_register_dry_run_latest.json
uv run --extra dev python scripts/profile_roar.py --iterations 3 --top 15

Results:

• ruff check .: passed
• mypy roar: passed
• pytest -m "not live_glaas and not ebpf": 1021 passed, 1 skipped
• focused cli_register_dry_run: 106.4ms
• full profiling artifacts refreshed under tests/benchmarks/results/

Follow-Up

The next highest-value work is deeper phase timing around roar run, plus runtime startup/atexit overhead behind the sitecustomize performance guardrail.