fix: unbreak the v1.2.0 release notes and tighten the fragment ceiling to 200

[skyoo2003]

Summary

Clears what stood between main and a v1.2.0 tag — six changelog fragments whose issue links render empty, which the release workflow rejects — then halves the fragment body ceiling to 200 characters and refits every unreleased fragment to it. Also corrects a protocol table in docs/coverage.md that miscounts EC2.

Related Issue

Refs #160, #161, #162, #163 — this corrects fragments and documentation those PRs added while still unreleased. No separate issue filed.

Changes

Release blocker — empty issue links

• Six fragments wrote Issue: at the top level instead of under custom:, where .changie.yaml declares it. changie batch v1.2.0 rendered them as [#<no value>](.../issues/<no value>), and release.yml:234 rejects any entry not ending in issues/<positive int>)). The tag would have failed at notes validation, after the Go suite, the boto3 suite and the codegen drift check had all passed.
• The pre-flight check in RELEASE.md does not catch this. grep -L 'Issue: "[0-9]' matches both the broken and the correct spelling, so the checklist passed on a tree that could not be tagged.

Fragment ceiling 400 → 200

• 400 was wide enough to hold two full sentences plus a subordinate clause, so it never pushed against the one-sentence rule it was meant to back — the #163 fragment reached 484 and still read as being within the spirit of the limit. At 200 the ceiling and the rule push the same way.
• All nine unreleased fragments now fit, longest 199. What came out is detail the linked issue already carries: the eight service names in feat: hand-write the eight services the CRUD engine cannot classify #161, the CI-runner timing in feat: state coverage as two targets and gate the three figures nothing read #163, the DynamoDB and Lambda operation counts in docs: stop the per-service pages promising the opposite of what the code does #167.
• RELEASE.md's own "right length" example was 203 characters and would have failed the limit stated in the paragraph above it. Trimmed to 193; both examples are now labelled with their length.

docs/coverage.md protocol table

• no in-tree model 12 → 11, unrecognised protocol 1 → 2. Counted against internal/generated/fidelity/manifest_gen.go: 11 entries carry ModelBacked: false, and 13 carry Protocol: "" — the two-service difference is ec2 (aws.protocols#ec2Query, per smithy-models/ec2.json) and partnercentralrevenuemeasurement (rpcv2Cbor). internal/codegen/parser.go:441-453 recognises neither.
• Prose corrected in three places: "Only the second still stops it" was false once two services are blocked by the first; "exactly one service" was two; and a new paragraph separates what EC2 loses (the engine fallback — its tail stays unimplemented instead of auto-crud) from what a registered-only service loses (EC2 is served by a hand-written provider and is not in the registered-only five).

Test Plan

• changie batch v1.2.0 --dry-run | grep -c '<no value>' → 0 (was 6); all nine entries render a live issue link
• Every fragment body ≤ 200 chars (longest 199, was 484)
• CGO_ENABLED=0 go test ./cmd/devcloud/ → ok
• hugo --gc --minify && ./scripts/check-site-links.py → 904 internal links across 24 pages all resolve
• Table rows sum to 431 both before and after, which is why the arithmetic never exposed the error — and why no test gates this table

Full pre-flight on this tree, for the record: codegen drift zero (regenerated to a temp dir and diff -rq against internal/generated), go test ./... green, make test-compat 1,527 passed / 2 skipped / 1 xfailed.

Checklist

• Self-reviewed the code
• Added/updated tests — N/A, but see Follow-up
• Lint/format passes (golangci-lint run) — pre-commit hooks passed on all three commits
• Updated documentation (if applicable)
• Added a Changie changelog fragment for user-facing changes — N/A: every fix here corrects unreleased work (feat: register the 218 scaffolded services and stop their routes fabricating successes #162 introduced the wrong table values, and the fragments have never been batched), so the v1.2.0 notes already describe the correct end state

Follow-up

docs/coverage.md states "Every figure on this page is asserted against the binary by go test ./cmd/devcloud/", but the protocol table is not among them — coverage_test.go contains no reference to it. That is how these two rows drifted. Gating the table's protocol distribution against the manifest is worth a separate PR.