Add GeoSkills v0.4 workflow core

[gronbow]

Summary

• add one provider-neutral local geoskills.py entry point with version, self-check, review-only plan, and guarded run commands
• add strict geoskills.recipe/v1 YAML recipes, a fixed REE/spider/Harker/TAS registry, content-addressed plans, and all-or-nothing multi-task output
• add shared input, analyte, validation, plotting, export, reporting, error, and workflow modules while retaining the v0.3 command-line contracts
• add publication-oriented full-frame figures, collision-checked legends, editable SVG/PDF, and 600 dpi PNG/TIFF validation
• add privacy-safe shareable reports, beginner-oriented Chinese guidance, examples, documentation, CI checks, and regression coverage

Why

The v0.3 plotting scripts were scientifically reviewed but remained separate entry points. This candidate provides one deterministic local workflow that makes column mapping, units, scientific confirmations, output scope, and user approval explicit before plotting. It also strengthens overwrite recovery, stale-plan protection, artifact validation, and privacy boundaries.

During independent audit, shareable Harker output was found to expose exact source extrema. The v0.4 candidate removes those values, rejects equivalent extrema fields at report validation time, and adds permanent regression tests.

User impact

Users can prepare one reviewable recipe and safely run one or more of the four already-supported diagram families without uploading their table. The release does not introduce a new scientific diagram or replace professional review of normalization references and TAS applicability/boundaries.

Validation

• 126 core/workflow/security tests passed
• 48 plotting tests passed
• 22 inspection/normalization tests passed
• 196 tests collected in total
• Skill structure validation: Skill is valid!
• git diff --check passed
• synthetic and published-data regression workflows passed
• independent visual QA and fresh-environment workflow audit passed
• checked shareable bundles contained no absolute paths, sample identifiers, exact source extrema, sensitive source-data CSVs, or tested exact raw values