Add [Operation] and align the code-first sample with its contracts

[ipjohnson]

What

[Operation("listTodos")] on a code-first handler declares the operationId the document publishes. The generator carries it on the handler model beside Tag, writes it as given, prefixes a derived name that collides with a declared one with its tag (as it already does for two derived names), and reports two handlers declaring one id as HRDOA004, an error. Spec-first handlers carry their description's id verbatim instead of a camel-cased round trip, so the Smithy fixture's exported document moves from getPet to GetPet, which is what Smithy's own conversion writes.

The template's code-first sample declares the contract's four ids and the contract's constraints, [Range(Min = 1)] on the id and [StringLength(1, 64)] on the title. Its document now declares the same 400s with the same RequestValidationError schema, the minimum and minLength/maxLength the contract has, and the same operation ids, so the Refit method names, the three constraint tests and the two malformed-id tests run in every contract mode with no codeFirst or specFirst branch. DocumentStatusTests keeps one expectation set for every mode that declares its statuses; code-first throws mode keeps its own.

The socket test's Kiota and Refit blocks gain the NUnit assert pair #284 put around every other assert. Merged after #284, #283 had left them xUnit-only, so the script's kestrel:code:response:kiota:nunit row does not compile on main.

Why

A code-first handler's id was the method name, so the client generated from a code-first service differed from the one generated from the same service written spec-first, and the template's tests had to know which way the service was written. The id is part of the contract: a Refit interface names its methods after it, and deriving it from the method name made a C# rename a breaking change on the wire. With the ids and the constraints declared, one client comes out whichever way the service was written, which is what lets one test file cover both.

Verified: the CI-parity build, the full suite (7266 tests, the Smithy fixture's export re-approved by the build and the public API re-approved for the attribute), six new generator tests for the id derivation and the diagnostic, and six template rows scaffolded from a pack of this tree with zero warnings and 15 tests passing each: kestrel code-first with Kiota, Kiota under NUnit, Refit, Refit in throws mode and no client, and kestrel openapi with Refit. scripts/verify-templates.sh was running when this opened; its result follows as a comment. The generator source could not be validated against Hardened.Amz locally, as Amz main does not yet build against Framework main.

🤖 Generated with Claude Code

[ipjohnson]

scripts/verify-templates.sh finished after this opened, exit 0: "templates verified". 14 hardened-web rows built and tested from the packed template, including the NUnit, Moq and FakeItEasy rows #284 added and the three Refit rows whose tests now call the contract's method names code-first, plus the three Amz rows against published Hardened.Amz. Host independence held for src/Sample, src/Sample.Client and tests/Sample.Tests. The smithy rows were skipped, as the pinned Smithy CLI is not on this machine's PATH.