feat(pm): add --kind to set-field to correct an issue's type label post-filing

[aleskalfas]

What

Adds --kind to the project-management set-field verb so an issue's type:* classification label can be corrected after filing, in the same validated batch as --priority / --workstream / --parent.

• On a kind-driven (Task) issue, --kind <value> swaps the type:* label and realigns the title prefix per classification.yaml's title_prefix_by_value (e.g. [Chore] → [Bug]), keeping label and prefix coupled through the same single source of truth create-issue uses (axis_labels.resolve_write + the shared prefix map — no duplicated tables).
• On epic / feature / umbrella, where the structural type fixes the kind to feature, a non-feature --kind is refused up front, before any mutation, conforming to the structural_restriction hard-reject in classification.yaml / [project-management:DEC-011-title-formats] rather than manufacturing the mismatch. --kind feature on those types is a clean no-op.
• Unknown values and set-field's existing validate-before-any-mutation / idempotent-on-re-run posture are unchanged; --kind composes with the other fields in one batch.

Why

Correcting a mis-filed type:* previously required a raw gh issue edit --add-label/--remove-label (off the validated path) plus a manual edit-issue --title to fix the prefix. Surfaced live: #402 was filed type:maintenance but a broken shipped verb is a defect; the relabel to type:bug had to go through raw gh.

Design note (critic-reviewed)

The first cut allowed --kind to swap the label on a non-task while leaving the structural prefix — a critic pass found that manufactures exactly the kind/structural mismatch DEC-011 declares a hard-reject, and is a live break: open-pr/validate-pr derive PR conv-type from the closing issue's type:* label, and a non-task owning issue is the closing issue of its integration→main PR (DEC-013), so an EPIC relabelled type:bug would drive its closure PR to fix(...) against an [EPIC] title and the reviewer gate would reject the natural feat. Reworked to refuse-on-mismatch, which conforms to the existing decision (no amendment).

Verification

• test_pm_set_field.py 33 pass; pm suite 1749 pass; full suite 4137 pass.
• pkit schemas validate (13 pass); pkit migrations check-diff --include-working-tree --base main clean (pure addition — new optional flag).
• Dry-run smokes (mutated nothing): Task --kind docs → type:bug → type:docs + [Bug] → [Docs]; EPIC --kind bug refused up front naming the rule; EPIC --kind feature accepted as no-op.

Version / migration

Capability surface change (new flag) → package.yaml 0.41.0 → 0.42.0 (minor). No migration (additive optional flag).

Doc impact

• .pkit/capabilities/project-management/README.md — set-field row, failure-list bullet, and example block state the Task-only scope and the epic/feature/umbrella refusal.

Test plan

• Task path: label swap + prefix realign; prefix-already-correct no-op; idempotent re-run.
• Refusal path: non-feature kind on epic/feature/umbrella refused before any mutation.
• --kind feature on a feature/epic accepted as no-op.
• Unknown value refused; combined --kind --priority --workstream batch.

Follow-up (out of scope, surfaced by the critic)

DEC-011's structural_restriction hard-reject is currently enforced nowhere in code (create-issue / validate-issue both skip the cross-check), so a mismatched issue still validates clean. This PR makes set-field conform; the broader enforcement gap (and a shared _lib consistency predicate the three callers want, COR-007) is a separate issue.

Closes #404

🤖 Generated with Claude Code

[aleskalfas]

Reviewer agent (local, reviewer): APPROVED

pm-conventions review of PR #407 against the project-management capability's schemas. All criteria pass; no warnings, no audit-bypass findings.

• Branch shape ([project-management:DEC-013-branch-and-pr-conventions], git-conventions.yaml branch-name): feat/404-add-kind-to-set-field matches <conv-type>/<issue-number>-<slug>.
• Branch-type / issue-type alignment (git-conventions.yaml branch-name rule 2 + classification.yaml pr_type_mapping): closing issue [Task] Add --kind to set-field to correct an issue's type label post-filing #404 is type:feature → feat; branch segment is feat.
• PR title Conventional Commits ([COR-008] + DEC-013): feat(pm): add --kind to set-field … parses as <type>(<scope>): <description> with feat in the accepted list.
• PR title type matches issue type (pr_type_mapping): type:feature → feat; the title uses feat.
• Issue classification complete ([project-management:DEC-012-classification-axes]): [Task] Add --kind to set-field to correct an issue's type label post-filing #404 carries type:feature, priority:Medium, and workstream:capabilities.
• Closing-issue link (git-conventions.yaml pr-body): body has Closes #404 and closingIssuesReferences is populated.
• ## Doc impact section present (git-conventions.yaml pr-body, hard-reject): present, with the README set-field row / failure-list / example update checked.
• ## Test plan section (DEC-031 placeholder rule): present with authored, checked items — no empty-skeleton hard-reject.
• Surface-change discipline ([COR-010]): pkit migrations check-diff --base main reports no migration-triggering changes; the package.yaml 0.41.0 → 0.42.0 bump is an additive new-flag minor with no renames/removals or schema_version change — no migration required.
• No-shared-files invariant ([COR-001]): the diff touches only capability-owned paths under .pkit/capabilities/project-management/ and tests/; no */core/ or otherwise core-owned files edited.