Downsize spec and plan scope when the resulting plan is too large

[rochecompaan]

Summary

Add a sizing step to Patchmill planning: when the resulting plan is too large,
downsize the spec and plan instead of carrying an oversized plan into
implementation.

Motivation

Large plans are hard to review, hard to implement reliably, and hard to merge.
A plan that overshoots a manageable size should trigger scope reduction, not
heroics.

Proposal

• Measure the size of the plan produced by planning (for example number of
  steps, files touched, or estimated change size).
• When the plan exceeds a defined threshold, feed that back into planning so
  the spec is narrowed or split into smaller pieces.
• The result is a smaller spec and plan that can be implemented and reviewed
  on its own, with the remaining scope tracked as follow-up work.

Open questions

• What size signals and thresholds define "too large"?
• Does downsizing mean narrowing the spec, splitting the work into multiple
  issues, or both?
• How does Patchmill track the deferred remainder?

Acceptance criteria

• Planning detects oversized plans against a defined threshold.
• Oversized plans are downsized before implementation, and the deferred scope
  is recorded.

[rochecompaan]

This was generated by AI during triage.

Triage Notes

What we have established so far:

• Planning should apply a size gate before implementation begins.
• If a generated spec/plan is oversized, Patchmill should reduce it to a coherent implementable slice rather than proceeding unchanged.
• Deferred scope must be recorded durably.

What we still need from you:

• Which deterministic size signal(s) and exact threshold(s) should v1 enforce (for example, implementation-task count, distinct files, or both)?
• Should downsizing rewrite the current spec/plan to one coherent slice, automatically create multiple follow-up issues, or only record suggested follow-ups for a maintainer to create?
• Where and in what minimum format should deferred scope be recorded, and must it link back to the original issue/artifacts?
• If the first downsized plan still exceeds the threshold, should planning iterate until it passes, stop after a fixed number of attempts, or hand off to a maintainer?