perry-ext-pg and perry-ext-mysql2 use plain spawn_blocking instead of spawn_blocking_with_reactor for socket I/O #10339

Problem

perry-ext-pg and perry-ext-mysql2 use plain spawn_blocking — not spawn_blocking_with_reactor — for real socket I/O. perry_ffi_async.rs:252-278 documents that plain spawn_blocking panics with "there is no reactor running" for exactly this kind of work, which is the stated reason spawn_blocking_with_reactor exists as a second shim (already used correctly by net/ws/http).

Why it matters

This evidently works today only because binding_needs_shared_tokio forces a shared tokio compilation for these bindings — but nothing in the code states that this indirect mechanism is what's actually making the unsafe-looking call safe. That makes it fragile: a change to binding_needs_shared_tokio's behavior, or to how these bindings are compiled, could silently reintroduce the "no reactor running" panic with no obvious connection to its cause.

What would fix it

Switch perry-ext-pg and perry-ext-mysql2 to spawn_blocking_with_reactor to match net/ws/http's already-correct pattern, removing the implicit dependency on binding_needs_shared_tokio's side effect — or, short of that, document explicitly at both call sites why the plain spawn_blocking call is safe here and what would break that invariant.