relay: subscription id length is measured three different ways #954

• decode (messages.rs): chars().count() — Unicode scalar values
• REQ/COUNT handlers (subscribe.rs): .len() — UTF-8 bytes
• TypeScript capture: UTF-16 units

With the shipped 256 limit, an id of 100 Chinese characters passes decoding (100) and is rejected by both handlers (300 bytes), although the capture accepts it.

Fix: encode_utf16().count() in all three, with non-ASCII cases in the tests.

Evidence: #948.