Report corpus and benchmark changes during regeneration

[andrew]

Corpus updates are deliberate commits, but reviewing their cost currently
requires several commands and manually copied results.

Add one regeneration report containing:

• Pinned ScanCode commit and corpus hash.
• Rule count and embedded size.
• Deterministic rebuild result.
• Cold New time, allocations, and allocated bytes.
• Repeated New cost and warm Match throughput and allocations.
• Conformance pass count and divergence changes.

Report cold startup separately from warm matching. A command-line scan pays the
startup cost, while a long-lived service normally pays it once. Keep the output
stable enough to compare with the previous checked-in corpus and include in a
corpus update pull request.