feat(blog): arity is the only check an exposed method gets

`@imqueue/validation` had zero coverage outside the package catalog and one
FAQ answer, and "zod" appeared in exactly one post. The nearest published
piece, /blog/type-safe-service-communication-typescript/, is entirely
compile-time — it says "fully typed" and "fails to compile" and never reaches
what happens once the argument is JSON on somebody else's queue. This is that
half.

Written against a real service on Redis 8.0.5 with a bare @imqueue/core caller
rather than against the README, which was worth it. Before an exposed method
runs, the only runtime check is IMQ_RPC_INVALID_ARGS_COUNT: a string, a number
and null all land in a parameter typed as a class, and undeclared properties
arrive intact. @classtype()/@Property() are not a runtime schema — they exist
for the generated client, on the calling side.

Four measured behaviours of @imqueue/validation carry the rest of the post:

  * the @validatable() buffer leak, which breaks a DIFFERENT class — a forgotten
    seal leaves `city` in the buffer and the next sealed class rejects valid
    credentials over a property it does not declare
  * transforming schemas validate and change nothing: z.coerce.number() accepts
    '3' and the body still gets the string, .default() fills nothing in and only
    makes the field optional
  * undeclared properties survive @validated(), because the parsed result is
    discarded; catchall(z.never()) is the explicit fix
  * a failure reaches a remote caller as IMQ_RPC_CALL_ERROR with the issue list
    as a message string, never as a ZodError — and the fix is not the decorator,
    whose throw happens outside the body, but parsing in the body and rethrowing
    with a code

Cost: 32 ns plain against 119 ns guarded, so 87 ns, against a 40 µs median local
round trip. The guarded and unguarded methods measure 39 µs and 40 µs over RPC.

Carries an ## FAQ section, so the post emits FAQPage markup for four questions.

npm test green on a clean build. Search KPI unmoved: P@1 macro 52.3% (identical
to the pre-post baseline), micro 61.1%, recall@6 90.1%.

Author: Mikhus