feat: per-repo label multipliers with fnmatch wildcard support (closes #1016)

[wdeveloper16]

Summary

Closes #1016.

Moves label-multiplier scoring from a single global table to per-repository configuration in RepositoryConfig. Each repo can now define its own label patterns (with fnmatch wildcard support) and a default multiplier, allowing custom label taxonomies like kind/feature, type:bug-fix, or release-prefixed labels such as 3.0/feature.

Changes

Core

• Remove global LABEL_MULTIPLIERS from constants.py
• Add label_multipliers: Optional[Dict[str, float]] and default_label_multiplier: float = 1.0 to RepositoryConfig
• Add resolve_label_multiplier(label, repo_config) helper using fnmatch for wildcard pattern matching
• JSON loading validates multiplier values in [0.0, 20.0] and enforces a max of 10 entries per repo, logging warnings and skipping invalid entries

Legacy scoring path

• PullRequest now stores current_labels: frozenset (all currently-applied labels, set at construction)
• New _resolve_label() in scoring.py: tries the last timeline-applied label first, falls back to the highest-multiplier fnmatch match across all current labels, then defaults to default_label_multiplier

Mirror scoring path

• _resolve_trusted_scoring_label and _calculate_pr_multipliers use per-repo fnmatch instead of the global table

Example config

"entrius/gittensor": {
  "weight": 1.0,
  "label_multipliers": {
    "kind/feature": 1.5,
    "type/*": 1.1,
    "*-dev": 0.5
  },
  "default_label_multiplier": 1.0
}

Tests

• All 17 original label-extraction test cases preserved, now driven through a per-repo config fixture that mirrors the old global table
• New parameterized tests for resolve_label_multiplier: exact matches, prefix/suffix/middle wildcards (kind/*, type:*, *-dev), case insensitivity, release-prefixed labels
• New tests for _resolve_label fallback logic: timeline candidate wins if it matches, falls back to highest-multiplier match from current labels, returns default multiplier when nothing matches
• Constraint enforcement tests via load_master_repo_weights: rejects dicts with >10 entries, skips individual entries outside [0.0, 20.0], resets out-of-range default_label_multiplier to 1.0, accepts valid configs end-to-end
• End-to-end wildcard matching tests: kind/*, type:*, *-dev, 3.0/*, multi-pattern with last-applied-wins semantics, no-match fallback to default
• Mirror label-resolution tests updated to use explicit label_multipliers on the config fixture instead of the removed global table

[wdeveloper16]

Hi, @anderdc
Ready to review.
I'd appreciate your feedback with this.
Thanks.

[wdeveloper16]

Ready to review.

[anderdc]

Duplicate of #1027 which is better scoped/implemented. Closing.