perf(formatter): use SmallVec for MemberChain collections

[mds-ant]

What

MemberChain::from_call_expression runs for every call expression in the file and allocates 3+ heap collections per chain: the initial member list, the VecDeque<MemberChainGroup> in the builder/tail, and a Vec<ChainMember> per group.

Change

Replace these with SmallVec<[_; 4]>. ChainMember is two pointers plus a one-byte enum, and chains/groups are typically 2–6 elements, so a 4-element inline buffer keeps the common case on the stack.

VecDeque was only used for one pop_front in maybe_merge_with_first_group. This feels a little heavy-handed; SmallVec::remove(0) over ≤4 groups is fine.

FormatMemberChainGroup gains a separate reference lifetime: SmallVec<[T; N]> (smallvec 1.x) is invariant over T, and the previous code relied on MemberChainGroup being covariant over 'b to unify the borrow lifetime with the inner 'b.

File	Before	After	Delta
pdf.mjs	15,308	4,655	−10,653 (−69.6%)
App.tsx	7,588	2,933	−4,655 (−61.3%)
antd.js	83,239	36,634	−46,605 (−56.0%)
kitchen-sink.tsx	20,458	10,986	−9,472 (−46.3%)
binder.ts	824	490	−334 (−40.5%)
checker.ts	18,642	11,947	−6,695 (−35.9%)
RadixUIAdoptionSection.jsx	54	54	±0 (0.0%)

Some performance numbers from perf stat -e instructions (median of 5 runs, release build):

File	Before	After	Delta
pdf.mjs	204,603,800	202,990,088	−1,613,712 (−0.79%)
App.tsx	110,011,608	109,345,945	−665,663 (−0.60%)
antd.js	1,607,371,443	1,600,715,018	−6,656,425 (−0.41%)

Verification

• cargo clippy -p oxc_formatter (+--features detect_code_removal) — clean
• cargo test -p oxc_formatter --test mod — 294 passed
• cargo run -p oxc_prettier_conformance — snapshots unchanged
• cargo allocs — snapshot updated

This PR was assisted by Claude Code.

[codspeed-hq]

Merging this PR will not alter performance

✅ 52 untouched benchmarks
⏩ 19 skipped benchmarks¹

---

_{Comparing mds-ant:perf/formatter-member-chain-smallvec (acef583) with main (6371fed)}

Footnotes

1. 19 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[leaysgur]

Sorry for the late review.
Thank you!