[DLANSAMA]

This PR has two commits.

1. Start 0.7.0.dev0 (docs/releasing.md step 6)

0.6.0 is published. The tag v0.6.0 points at b0ef746, and release.yml passed every step: the CI matrix, the build, the PyPI publish, and the GitHub Release.

I installed it from PyPI in clean Python 3.12 venvs:

• platecli==0.6.0 and platecli[tui]==0.6.0 (with textual 8.2.8) both installed.
• In both, plate --version printed plate 0.6.0 and plate --sim status exited 0.

Main now goes back to 0.7.0.dev0, so a source checkout never reports itself as the release.

Plan changes:

• L-2 (cut 0.6.0) and D-3 are retired.
• D-3's naming ruling goes in the new docs/plan/settled.md. "Bambu" is fine wherever it describes the printers (keywords, "for Bambu Lab printers"). It never goes in the product's name, "unofficial" or not.

2. Run CI only when there is something new to test

The repo is public, so its hosted runners aren't billed. Most runs were still repeats:

• 14 of the last 15 nightly mutation runs tested the same commit.
• Each merge re-ran the full matrix on files the PR had just passed.

What changes:

• Draft PRs skip the test matrix. It runs once, when the PR is marked ready (ready_for_review is now a trigger). Lint still runs on drafts. A skipped job reports success under its fixed name, and "Lint (ruff + bandit)" is a required check, so a skipped lint could sit beside a real failure on the same commit. A skipped matrix job reports its unexpanded name, so a draft never shows the six required checks as passed, and a draft can't be merged anyway.
• A push to main skips test and lint when the squash merge's tree is identical to the PR head it came from.
  • The ruleset only lets a PR merge after all six required checks pass on that head, so that tree has already been tested.
  • A merge onto a main that moved has a different tree and is still tested.
  • The new merged job uses git alone, because release.yml grants only contents: read.
  • It only looks at branch pushes. When release.yml calls this workflow from a tag push, the event is push with a tag ref, so the release gate always runs the full matrix.
  • Every error path falls through to testing.
  • Dry run against the last 8 merges: 4 would have skipped (chore: release 0.6.0 #154, fix: audit round 4 — at-most-once printer commands, pin precedence, heater limits, terminal-safe output #153, ci: bump astral-sh/setup-uv from 10.1.0 to 10.2.0 in the actions group #143, fix: audit round 3 -- printer safety, slicing correctness, setup #142), and 4 would still test, because main had moved under them (docs(agents): drop a dead module row, a wrong PR citation and change-history notes #147, deps: bump the python group across 1 directory with 3 updates #145, ci: bump the actions group across 1 directory with 2 updates #139, deps: bump the python group across 1 directory with 4 updates #141).
• The nightly mutation run is skipped when main's commit already has a green baseline. A manual run always runs. A failed baseline keeps running nightly until it passes. If the check itself fails, the run goes ahead.
• The weekly scheduled CI run is unchanged. It is the drift detector for firmware, OrcaSlicer, Printables and CVEs.

CONTRIBUTING.md gains a "Pull requests and CI" section. The README, docs/mutation-baseline.md and docs/releasing.md now describe the new behaviour.

Checked

• actionlint passes on all three workflows.
• ci_workflow_smoke and release_readiness_smoke pass.
• Full local gates on the version bump: ruff, ruff format, mypy, bandit, all smokes, build and package smoke, twine check, and 2217 tests at 93.53% coverage.
• An independent review of the workflow change found two problems, both fixed here:
  • Lint was being skipped on drafts.
  • The tree comparison could treat two failed lookups as "equal".

This PR was opened as a draft, so its own first run tests the draft path. Marking it ready tests the full matrix.