Allow Rust callers to omit the automatic post-resume skill reload #2827

Problem

The Rust SDK's shared ordinary/prepared resume path awaits session.skills.reload after a successful session.resume response. Returned reload errors are logged and tolerated, but an unanswered request keeps resume pending. A caller that already owns skill reconciliation cannot currently omit this additional SDK-issued operation.

This mechanism is reproducible with a framed in-memory transport: acknowledge session.resume, then observe the automatic session.skills.reload request and withhold its response. No live runtime or production incident evidence is needed for this SDK-level reproduction.

Proposed behavior

Add a Rust-local ResumeSessionConfig::with_reload_skills(bool) option. Unset or true must retain existing awaited, best-effort behavior; explicit false must omit the automatic request entirely, without a timeout, detached task, or retry.

Keep mandatory resume validation, MCP-auth interest registration, options readiness, event/callback ownership, and cleanup unchanged. The option must not be serialized to runtime or AHP settings. It must not disable skills, imply catalog or disabled-preference freshness, gate continued/shared work, or cancel an already-issued operation. Explicit caller-managed session.rpc().skills().reload() must remain available.

SDK regression coverage

Use synthetic framed controls for ordinary/prepared omission, retained default/true settlement and returned errors, mandatory readiness, event/hook/provider callbacks, failure/cancellation/connection-loss cleanup, and unchanged wire configuration.