fix(orchestrate): clarify model and endpoint profile parameters

[Tryanks]

Passing profile=gpt-6-sol to dispatch failed while the error advertised enabled profiles: gpt-6-sol. That list actually contained model names: profile selects a provider endpoint configuration. The failure occurred before effort validation, so retrying without both profile and effort incorrectly suggested that high effort was unsupported.

Clarify model and endpoint fields in the dispatch/collaborate schemas and injected configuration. Matching errors now label each model/endpoint combination, tell callers to omit profile for built-in endpoints, and explain that effort has not yet been validated. Endpoint selection and effort validation remain owned by the existing resolver.

The regression test reproduces all three rejected parameter combinations and verifies that model-only dispatch works with both default effort and explicit high effort. Its diagnostic assertions failed before the fix and pass afterward; existing named-endpoint coverage remains in place.

Validation on macOS:

• cargo fmt --all --check
• cargo clippy --workspace --all-targets --locked -- -D warnings
• cargo test --workspace --locked
• cargo-machete . (the local cargo machete invocation misrouted its subcommand as a directory)

No live provider calls were needed for the resolver reproduction. Other desktop platforms and iOS/Android/Web targets are covered by CI before merge.