fix(sdk): M1 SDK sweep fixes - macro hardening, doc fidelity

[mfw78]

What

Rewrites docs/migration/0.1-to-0.2.md section 7 to match the SDK surface that actually shipped (0.2 ships host traits, typed errors, and the single #[nexum_sdk::module] attribute; 0.1 never had an SDK crate at all), and adds an explicit note that Signer/Messaging/RemoteStore/provider()/#[shepherd::module] never shipped. Retargets doc 05's cross-link text from "rename table" to "what changed".

Hardens the #[nexum_sdk::module] macro (crates/nexum-macros/src/lib.rs) to reject, each with a targeted compile error, any on_-prefixed method not in the HANDLERS set (previously a silent no-op for typo'd handlers), trait impls, generic impls, and impls with zero recognised handlers.

Qualifies doc 05's testing-story claim to the SDK side of things, documenting nexum-runtime's feature-gated test_utils::TestRuntime as the separate in-tree component-level harness that sits outside the module-author SDK contract. Also corrects doc 05's CowApiHost parenthetical (submit_order, cow_api_request), the crate trees, and the first #[nexum::module] mention to name the real nexum_sdk path.

Adds an "Authoring a module" section to docs/sdk.md documenting the #[nexum_sdk::module] attribute and its handler set, and includes nexum-macros in the rustdoc command.

Why

Stacked QA fixes for the M1 SDK sweep (part of #136): the migration doc and doc 05 had drifted from what actually shipped, and the module macro silently swallowed typo'd or malformed handler methods instead of failing at compile time. Fixing both keeps the docs trustworthy for module authors and closes a real correctness gap in the macro. This PR is stacked on docs/m1-sdk-design-rewrite (its base branch).

Testing

All run inside nix develop --command, with --locked and --config 'build.target-dir="/code/nxm/runtime/target"':

• cargo fmt --all -- --check
• cargo clippy --workspace --all-targets --all-features -- -D warnings
• RUSTDOCFLAGS=-D warnings cargo doc --workspace --no-deps
• cargo build --release --target wasm32-wasip2 for example, twap-monitor, ethflow-watcher, price-alert, balance-tracker, stop-loss, flaky-bomb, fuel-bomb, memory-bomb
• cargo test --workspace --all-features --no-fail-fast
• Probe-verified the macro's new compile errors for a typo'd handler, a trait impl, a generic impl, and a zero-handler impl

AI Assistance

Claude (Opus) used for the implementation.

[lgahdl]

Good hardening. Three items below.

[lgahdl]

The second corollary may be inaccurate: the macro's own quote! output already uses ::core::result::Result, ::std::vec::Vec, and ::core::result::Result::Ok — fully qualified paths — so shadowing those names in the crate root would not break the macro-emitted code. The constraint, if real, comes from the wit_bindgen::generate! expansion itself, not from "wit-bindgen's generated Guest trait" broadly. Worth verifying which names generate! actually emits unqualified before documenting this as a constraint; if it turns out the expansion is also hygienic, the corollary can be dropped.

Suggested change

	/// namespace), and the crate root must not shadow std prelude names
	/// such as `Result`, `Vec`, or `Ok` (wit-bindgen's generated `Guest`
	/// trait refers to them unqualified).
	#[proc_macro_attribute]
	/// `CARGO_MANIFEST_DIR`. One corollary: the consuming crate must
	/// declare `wit-bindgen` as a direct dependency (the emitted
	/// `wit_bindgen::generate!` call resolves against the consumer's
	/// namespace, not `nexum-macros`'s).

[lgahdl]

The error tells the user to rename, but the lower-friction fix is to move the helper to a separate impl block — the macro only inspects the single attributed impl. A user with a well-named on_retry_cooldown_expired helper would prefer that over renaming.

Suggested change

	format!(
	"`{name}` is not a recognised #[nexum_sdk::module] handler; expected one \
	of {HANDLERS:?} (rename helpers so they do not start with `on_`)"
	),
	format!(
	"`{name}` is not a recognised #[nexum_sdk::module] handler; \
	expected one of {HANDLERS:?}. Move helpers that start with \
	`on_` to a separate `impl` block, or rename them."
	),

[lgahdl]

"Earlier drafts" is ambiguous — it reads as earlier commits of this file (stale docs) rather than features that were planned and cut. A reader who wrote code against those APIs gets a different message from each reading.

Suggested change

	Earlier drafts of this section described a richer 0.2 surface
	This section previously described a richer 0.2 surface