feat(workflow): mastermind-style-deep writes a developer portrait, not count-bullets

[aglumova]

Rewrites the mastermind-style-deep skill from a flat list of grepped counts into a grounded developer portrait.

Why

The first version produced 8 isolated count-bullets — useful, but it (a) asserted preferences like "iterators over loops" without counting the alternative, and (b) credited things a linter forces (e.g. /// on every fn under #![deny(missing_docs)]) as personal signature. It read like a lint report, not a description of a person.

What

The skill now instructs the agent to write a portrait organized by the dimensions of how someone actually develops: design & problem-solving, code shape, comments (do they write them at all?), tests (parametrized or one-off?), optimization (measured or not?), what they pay attention to, and commit voice. Prose per dimension, every claim tied to a tell.

New guards, all addressing real gaps:

• Count the contrast — "X over Y" needs both numbers (e.g. 904 iterator calls vs 185 for-loops).
• Check enforcing config first — rustfmt.toml / #![deny(missing_docs)] / clippy.toml / CI lint steps; a tool-forced trait is a measured rule, not a signature, and is excluded.
• Read 4–6 real files, not just grep — design/optimization/ergonomics intent isn't visible in counts.
• Negative space is signal — "no property tests", "doesn't micro-optimize" belong in the portrait.

Testing

python scripts/validate.py clean (20 artifacts). Demonstrated on the mmcg crate — the portrait correctly distinguished personal habits (heavy doc-comments, no missing_docs lint) from enforced ones (clippy -D warnings in CI), and grounded every dimension in counts.