Add Python repository graph extractor

[RomanEmreis]

What problem does this solve?

The generic extractor exposes Python files, but Ferrus cannot navigate packages, modules, declarations, imports, or inheritance as structured graph facts.

Proposed solution

Add a deterministic Python extractor through the existing Extractor boundary.

Scope:

• Support .py and .pyi files, package directories, and namespace-package layouts.
• Emit modules, classes, functions, async functions, methods, properties, type aliases, and bounded declaration metadata with evidence spans.
• Emit import, from ... import, relative import, re-export, and statically explicit inheritance relationships.
• Resolve repository-local modules and packages conservatively across __init__.py, namespace packages, and supported source-root declarations.
• Parse relevant static pyproject.toml package metadata without importing modules, running Python, loading plugins, or invoking build backends.
• Preserve dynamic imports, wildcard ambiguity, runtime path mutation, and unsupported targets as unresolved evidence.

Acceptance criteria

• Identical immutable input produces deterministic node, edge, semantic-key, and diagnostic output.
• Incomplete or malformed source produces bounded diagnostics while preserving recoverable facts.
• Parser time, file, fact, value, and diagnostic limits are enforced during extraction.
• Source bodies, default values, secrets, and absolute local paths are not persisted.
• Add/change/delete/rename and package-layout fixtures invalidate only the required fragments and resolution facts.
• Tests cover regular and namespace packages, relative imports, stubs, async declarations, re-exports, inheritance, and unresolved dynamic imports.
• User documentation lists supported syntax and known resolution limits.

Impact on workflow

None. This extends optional derived repository context and must not alter task, run, lease, review, or approval state.