[Epic] Client-directed skill selection

## Client-directed skill selection

### Problem

An Agent Card can advertise multiple skills, each with its own `id`, but a client sending a message has no standard way to indicate *which* advertised skill it intends to invoke. For multi-skill agents — especially when an orchestrator already knows the target skill — the routing is left implicit.

### Consolidated issues

- [ ] #143 — Task submission should provide a way to specify the requested skill(s) *(anchor)*
- [ ] #1812 — Add an optional `skill_id` routing hint to `SendMessageConfiguration` *(concrete proposal for #143)*
- [ ] #883 — Managing multi-step user interactions and skill routing
- [ ] #902 — Skill selection and multi-step interaction gaps
- [ ] #1776 — Conventions for Schema/Mode Negotiation and SDK Level Type Dispatch for Media Types

### Acceptance criteria

- An **optional** skill identifier/hint can be supplied on send (e.g. on `SendMessageConfiguration`).
- Semantics defined for unknown/unsupported skill id (error vs. ignore).
- Strictly additive — single-skill agents and existing clients are unaffected.
- Reflected in `a2a.proto` and all three bindings.

The issue references a2a.proto and three bindings; start by examining the SendMessageConfiguration definition there. Review the linked issues, especially #1812, for context on the concrete proposal. The work involves updating the protocol specification and ensuring changes are additive and backward compatible. Testing will require understanding how multi-skill agents currently handle routing.

Scritto dal modello di indicizzazione a partire dal testo della issue.