feat: `dbts sql` — one-shot or interactive SQL on the sandbox

[luiul]

Summary

Run SQL against the sandbox without leaving the CLI. dbts sql -q "SELECT ..." for one-shot queries (Rich-rendered table); bare dbts sql opens a REPL.

Why

The most common manual context-switch in the dev loop today: after every dbts build, you SELECT * to sanity-check the result — currently means switching to Snowsight or maintaining a separate snowsql config. dbts sql already has the right target, role, and warehouse.

Sketch

• dbts sql -q "..." — execute, render rows as a Rich table, exit.
• dbts sql -f file.sql — read query from file.
• dbts sql — REPL (multiline, history, semicolon-terminated). Either shell out to snowsql if installed, or use prompt_toolkit for a small interactive loop.
• --target flag like the dbt pass-throughs (default sandbox).
• --limit N (default 100) so accidental SELECT * from a billion-row table doesn't hang the terminal.

Where it'd live

• New module src/dbts/sql_cmd.py (avoid shadowing stdlib sql).
• Reuse snowflake.connect / run_sql.
• Output uses rich.table.Table (same as clone.status).

Effort

Small for the one-shot path, medium for the REPL.

Tier

Tier 1 — recommended (highest-frequency missing dev action).