Add nx_log! usage examples and unit tests for the log macro

[lntutor]

Summary

• document literal, formatted, and multi-argument nx_log! usage
• capture log messages in unit tests without invoking the wasm host ABI
• assert the exact output produced by each formatting form

Verification

• /Users/loi/.cargo/bin/cargo test -p nx-sdk
• /Users/loi/.cargo/bin/cargo doc -p nx-sdk --no-deps
• /Users/loi/.cargo/bin/cargo fmt --all -- --check
• /Users/loi/.cargo/bin/cargo clippy -p nx-sdk --all-targets -- -D warnings
• git diff --check

Closes #67

[lntutor]

Current head de6cac0d0403099a16b0f63deb80b4769f95ad6c is mergeable and the visible CI surface on this exact SHA is fully green: cross-platform Check, Clippy, and Test jobs on Ubuntu, macOS, and Windows; Format; Build WASM Examples; CLI Multi-process Smoke; Benchmark Comparator Tests; Ubuntu Profiling Artifacts; Benchmark Regression Gate; and the aggregate CI Success check all completed successfully. There are no review threads or review-body comments open on the current PR state, so if this nx_log formatting test coverage still matches the intended direction, it looks ready for maintainer review/merge when convenient.

[GianIac]

Thanks for this PR.
Good coverage of the three formatting patterns and the doctest gating with #[cfg(target_arch = "wasm32")] is a nice touch.

• PR has conflicts with main (likely from #90 adding #[inline] to log::log). Please rebase and keep the #[inline] on the #[cfg(not(test))] version.

Question

• The issue asked for "docs + tests, no public API changes", but this replaces log::log with a fully separate #[cfg(test)] in-memory implementation. It works, but it's more invasive than needed and cascades into the #[expect(dead_code)] on host_log_v2.
• Would you consider a lighter approach? e.g. testing only the macro expansion via a small #[cfg(test)] fn fmt_message(args: fmt::Arguments) -> String and asserting on that, without swapping out log::log itself. That keeps log.rs single-implementation and closer to the issue scope.
• If you prefer to keep the current approach, that's OK too just add a short comment at the top of log.rs explaining why there are two log::log definitions.

Nits (optional)

• Consider pub(crate) on take_test_messages so future modules can assert on emitted logs.
• The assert_logged_message helper is nice, worth keeping either way.

Once rebased and we've agreed on the design direction, happy to merge.