Exclude vendored and generated files from git line-count metrics

[aleksdotbar]

Problem

Git line-count metrics (e.g. "Code lines added", "Lines added") sum every
added line reported by the commit diff. Files are bucketed only by
extension (code / docs / config), with no consideration of whether the
file is authored or machine-produced. As a result, content that a person did
not write by hand is counted as their contribution:

• vendored dependencies checked into the repo (e.g. .venv / site-packages,
  node_modules, vendor/)
• build/generated output (e.g. dist/, build/, minified bundles, generated
  code or data files)
• lockfiles and other machine-managed artifacts

A single commit that adds such a directory can contribute hundreds of
thousands of lines under a code extension (e.g. vendored .py/.js), which
are then attributed as authored code.

Impact

• Per-person distortion: individuals whose repos happen to include
  vendored/generated content show wildly inflated line counts that do not
  reflect real output. This correlates with role and repo hygiene rather than
  contribution.
• Peer-pool distortion: because peer standing is computed against a
  cohort (e.g. department) median/quartiles, one such outlier skews the
  benchmark for everyone in that cohort, degrading every peer comparison in
  the group.
• Line-count metrics are currently untrustworthy wherever a repo does not
  perfectly gitignore generated/vendored content — which is common.

Root cause

File classification keys on file extension only. There is no path-based
exclusion for vendored/generated content, so such files pass through the same
as authored source.

Proposed direction

Add a path-based exclusion applied before line-count aggregation, so
vendored/generated files are not counted toward authored line metrics:

• Maintain an exclusion list of well-known vendored/generated path patterns
  (dependency directories, build output, minified assets, lockfiles). Treat it
  as configuration so it can evolve without code changes.
• Apply the exclusion in the git file-change classification / observation
  layer so it flows consistently to every downstream line-count metric and to
  peer aggregation.
• Consider a separate "generated/vendored" category rather than a hard drop,
  so the signal is auditable and could be surfaced distinctly later.

Acceptance criteria

• Vendored/generated files are excluded from authored line-count metrics.
• Exclusion patterns are configurable, not hard-coded per call site.
• Exclusion is applied once, upstream, and reflected in every line-count
  metric and in peer/cohort aggregation.
• Coverage for representative vendored/generated path patterns.

Scope

Independent of the metric-family migrations; affects the git domain only.