chore: Prepare v1.7.0 release

[keithdv]

Closes the TRIM arc. Version 1.6.1 → 1.7.0, release notes authored, release-step obligations from deferred row 22 discharged.

What v1.7.0 ships

Under-preservation (types trimmed away that shouldn't be) — positional-record DTOs in factory signatures, DTOs carried as [Factory] entity properties, and FactoryEventBase descendants reachable only from a subscription lambda. Plus restored incremental caching on the relay-handler branch.

Over-preservation ([Remote] bodies shipping to the browser) — two independent mechanisms, both fixed:

Mechanism	Affected since	Shapes
Registrar attribute's [DynamicallyAccessedMembers] named the consumer's own class	v0.21.2	[Execute] static factories, [FactoryEventHandler<T>] classes
IsServerRuntime guard does not fold inside an async state machine	always	every class factory with an async operation, LocalSave, class-level [Execute]

What the release notes carry that commit-prefix scanning could not

CLAUDE.md's documented process scans commit prefixes. Run literally it would have produced three feat: lines, one perf:, three fix: — and omitted the behaviour change, because TRIM-009's commit is fix:-prefixed and prefix-scanning cannot know a bug fix changed observable semantics. Deferred row 22 existed for exactly this.

• Behaviour Changes — the server-only guard now throws synchronously rather than surfacing as a faulted Task. Awaiting callers are unaffected; migration guidance covers the split-call/await and Task.WhenAll shapes, which are the only ones that observe it.
• What this release does not claim — the interface-factory carve-out, stated in the consumer-facing artifact rather than only in the todo, the generator comments, and the skill. That leg has received neither fix and is structurally unmeasurable from a client-side harness; body removal there is not established.

Also at the release step, per recorded dispositions

• TrimmingTests/README.md "How It Works" rewritten (audit callout C8) — it described only the pre-TRIM-009 mechanism. Holder-prefix table taken from the renderers by grep; the first draft named the relay prefix from memory and got it wrong, which is recorded.
• ClassFactoryWithExecute.cs gains the trimming commentary block (deferred item 11, TRIM-009 half). The TRIM-002/007 half stays accepted-with-reason — preservation genuinely is unobservable in untrimmed Design tests; TRIM-009's behaviour change is not.
• todo.md's Close-Out Audit section now links the 2026-08-14 whole-arc audit, which it never did — the arc's final review was undiscoverable from its own container.
• 1.x nav_order chain renumbered to a strict 1–9 rather than adding a fresh collision at the top. The v0.x block stays as declared out of scope in d8369b3; now tracked as deferred row 24.

Verification

• No public API changed — no attribute, type, or method signature added, removed, or altered since v1.6.1. Library diff is XML-doc only.
• Design.sln build: succeeded, 0 warnings, 0 errors (the new XML doc block compiles under TreatWarningsAsErrors).
• Neatoo.RemoteFactory.sln build: succeeded, 0 errors. 4 warnings, all pre-existing and untouched by this change (WASM NativeFileReference in an example project; two CA1062 in TrimmingTests).
• No test asserts a version string — every 1.6.1/1.7.0 hit outside Directory.Build.props is a generated obj/ AssemblyInfo.cs.
• Full suite runs here in CI.

Acceptance

AC1–AC4 and AC6 met. AC5 is discharged without consumer proof, per its close-out restatement (audit C4) — the container states this plainly rather than ticking it. The LinkerConfig deletion and Release WASM publish belong to zTreatment PCB-003, now unblocked.

After merge: tag v1.7.0 off main, CI publishes to NuGet and creates the GitHub release.

🤖 Generated with Claude Code