feat: Make ACCEPTS header parsing follow specification more closely

[kibertoad]

Per RFC 9110 §12.5.1 (https://www.rfc-editor.org/rfc/rfc9110#name-accept), a wildcard Accept header (/, text/*) or a missing Accept header means any media type is acceptable, including text/event-stream. The server gets to choose. @fastify/sse currently treats those headers as "not SSE" and falls through to the route's regular handler. For SSE-only routes the regular handler is the same handler that uses reply.sse, and reply.sse is undefined on the fallback path, so the request 500s with a cryptic TypeError: Cannot read properties of undefined (reading 'send').

That hits anything that doesn't send an explicit text/event-stream header: Postman defaults, curl defaults, fetch() defaults, plain navigations.

After this PR, routes can now declare their intent up front:

• sse: 'only' - SSE-only route. Lenient gate per RFC: any spec-compliant Accept (text/event-stream, text/*, */*, missing) admits SSE. Clients that explicitly refuse SSE (Accept: application/json, or text/event-stream;q=0) get 406 Not Acceptable.
• sse: 'dual' - Same handler serves both SSE and non-SSE clients. Strict gate: only an explicit text/event-stream token routes to SSE; everything else (including /) reaches the handler with reply.sse undefined, so the handler can serve a non-SSE response. This preserves the convention that wildcards prefer the richer non-SSE representation when the route offers a choice.
• sse: true - Unchanged routing (back-compat). On the fallback path the plugin now rethrows with a clearer error if the handler trips on undefined reply.sse, naming sse: 'only' as the fix.

sse: true routes behave exactly as before - the existing fallback pattern, the existing tests, and the documented examples all keep working. New routes can opt into the explicit kinds without affecting old ones.

Safe for a minor bump pre-1.0: no behavioral change for any existing sse: true registration; only additive surface ('only', 'dual', the object form's kind field).

Checklist

• run npm run test && npm run benchmark --if-present
• tests and/or benchmarks are included
• documentation is changed or added
• commit message and code follows the Developer's Certification of Origin and the Code of conduct

[kibertoad]

@mcollina do you think optimization was sufficient?

[kibertoad]

@mcollina could you please rereview? hot path is now faster than it was before the PR

[kibertoad]

@mcollina how can we proceed with that? or you prefer not to do this change altogether?

[mcollina]

lgtm

Sorry, I lost all the notifications

[andokai]

Hi, any sign of a release with these changes included?