Lock-free root fast paths for hygiene queries [-0.06%, below significance]

[xmakro]

outer_expn_is_descendant_of and normalize_to_macros_2_0/_macro_rules took the HygieneData lock (thread-local + RefCell, or a real mutex under the parallel front end) to compute answers that are statically known for the root syntax context — the common case for non-macro code, hit several times per name resolution.

Correctness argument, verified exhaustively: the root context's data is written once in HygieneData::new and never mutated afterwards. alloc_ctxt can never reserve index 0 (the vec starts with one element), update_dollar_crate_names writes only dollar_crate_name, and both the metadata and on-disk-cache decoders short-circuit raw_id == 0 before allocating. is_descendant_of(_, root) is unconditionally true — the existing sibling fast path checks ancestor == root before its cross-crate guard, so foreign expansions are correctly descendants.

Honest measurement: -0.06% geomean, zero cells clearing 0.25%. Real but below what a perf run would flag as significant. Submitted because it is provably equivalent and removes lock traffic that matters more under -Zthreads>1, not because the number justifies it on its own.

Known trade-off: a nested HygieneData::with bug would now abort only when the context happens to be non-root, rather than deterministically.

Measured in isolation against a clean base (ThinLTO stage2, glibc, instructions:u, full scenario, 5 crates x 3 profiles = 15 cells). This change was measured alone, not as part of a batch.

Reviewed adversarially before submission; findings that survived are reflected in the code and described above.