Gravity Flow: run the bench release gate for gravityflow_* tools

[marceldarvas]

Run the small-model bench (npm run bench) covering the gravityflow_* tools before tagging a release. A small model failing means the tool surface (descriptions/schemas/errors) is too hard for an agent — fix the surface, not the gate.

To do

• Add/confirm bench task coverage for Gravity Flow workflow actions.
• Run npm run bench against a site with Gravity Flow installed (see AGENTS.md "Bench" — use --mint or pin local GRAVITY_FORMS_*; BENCH_PLUGINS to include gravityflow).
• Address any sub-threshold tasks by improving tool descriptions/schemas.

Branch: feat/gravity-flow