feat(metrics): directory- and repo-level bus factor (Avelino DoA)

[dekobon]

Follow-up to #328.

Add directory- and repo-level bus / truck factor as a vcs aggregate, complementing the v1 file-level ownership_top_share.

Why

Bus factor (a.k.a. truck factor) measures the minimum number of developers whose departure would leave the project (or a subsystem) without sufficient knowledge to continue. Avelino et al. propose a Degree-of-Authorship (DoA) heuristic that iteratively removes the most knowledgeable developer until >50% of files are abandoned (see https://www.sciencedirect.com/science/article/pii/S0020025526002847).

The v1 file-level ownership_top_share captures concentration per file. Bus factor captures concentration across a set of files, which is the directionally correct unit for component- or repo-level risk planning.

Scope

• New vcs::BusFactor computation:
  • Per-directory bus factor (each top-level directory and each src/ subdirectory)
  • Repo-level bus factor
• Avelino DoA formula: DoA(d, f) = N₁ * FirstAuthorship + N₂ * ln(1 + Deliveries) + N₃ * ln(1 + AcceptedChanges) with thresholds from the paper.
• Surface in a new top-level vcs_aggregate object alongside the per-file vcs data.

Edge cases

• Files with single authors: skew bus factor downward; document.
• Bots (already filtered at file level) must also be filtered here.
• Very large directories: cap iteration count.

Acceptance criteria

• vcs_aggregate.bus_factor.{repo, by_directory} emitted alongside per-file vcs data.
• Configurable threshold (default 0.5, per Avelino).
• Unit tests on synthetic repos with known authorship distributions.