feat(cli,db,render): add row filters, cost sorting, run-rate projection, and CSV output

[lorecanc]

Summary

Additive output/analysis features for run:

• Row filters — --provider, --model, --agent (all repeatable) and --exclude-provider. Implemented as a Filters dataclass + _filter_clause() applied to every query (_base_query, by_agent, by_session, totals), parameterized SQL only.
• --sort tokens|cost|calls — sorts grouped rows by cost or calls instead of always tokens.
• Run-rate projection — the summary panel shows ≈ $X/mo (cost projected onto a 30-day month at the observed pace); also emitted as run_rate_monthly in JSON.
• --csv — row export to stdout (mutually exclusive with --json), with the agent view including the model column.

Cross-PR note

The Filters dataclass is identical to the one introduced by the granularity PR (feat/time-granularities) so the hunks deduplicate at merge; the granularity PR's by_time() already forwards a reserved filters parameter.

Changes

• src/opencode_usage/db.py: Filters, _filter_clause(), filter/sort params on _base_query, daily, by_model, by_provider, by_agent, by_session, totals
• src/opencode_usage/cli.py: filter/sort/csv flags, _build_filters, _compute_run_rate, JSON run_rate_monthly
• src/opencode_usage/render.py: run-rate in render_summary, render_csv
• tests/test_db.py, tests/test_cli.py: new tests (filters per query, exact-ID matching, sort order, flag parsing, exclusivity, run-rate math)
• README.md: documented

Verification

• uv run pytest tests/ — 390 passed
• uvx ruff check . and uvx ruff format --check . — clean