Segfault regression at 2.1.242 on glibc 2.44 (Arch) — distinct from closed #62747

### Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported as a fresh regression
- [x] This is a single bug report
- [x] I am using the latest version of Claude Code

### What's Wrong?
Claude Code segfaults immediately on startup (`claude --version` never prints, exit 139) on very recent glibc. This looks related to #62747, but that issue was closed as fixed in 2.1.197, and the regression window I bisected is completely different — suggesting either a reintroduced regression or a distinct bug with an identical symptom.

### Environment
- Distro: Arch Linux (rolling release)
- Kernel: 7.1.9-arch1-2
- glibc: 2.44
- CPU: Intel i5-11260H (full AVX2/AVX-512 — ruling out the separate CPU-feature crash reported elsewhere)
- Install method: npm (`@anthropic-ai/claude-code`)

### Steps to Reproduce
1. `npm install -g --allow-scripts=@anthropic-ai/claude-code @anthropic-ai/claude-code@2.1.242`
2. `claude --version`
3. Segfaults immediately, no output

### Bisection
Tested each version explicitly via npm install + `claude --version`:
- ✅ 2.1.212 — works
- ✅ 2.1.228 — works
- ✅ 2.1.235 — works
- ✅ 2.1.239 — works
- ✅ 2.1.241 — works
- ❌ 2.1.242 — segfaults
- ❌ 2.1.243 — segfaults

### Is this a regression?
Yes — exact boundary is 2.1.241 (works) → 2.1.242 (broken).

### Related issues
- #62747 — same symptom (immediate SIGSEGV, Bun runtime), closed as fixed in 2.1.197. A separate user (@nic116seitz) commented there recently on Arch/glibc 2.44 reporting the crash persists even on 2.1.205, which is inconsistent with that fix. This report's own bisection (2.1.241→2.1.242) is a third, distinct regression window from either of those, which is why I'm filing separately rather than adding a fourth comment to a closed thread.

Happy to test any candidate fix or provide a coredump/backtrace if useful.

Start by reproducing the failure with the listed npm install command and `claude --version` on Arch Linux with glibc 2.44. Compare 2.1.241 and 2.1.242, then collect the coredump or backtrace requested in the report. Done means startup succeeds without SIGSEGV on the affected environment.

索引モデルが issue の本文から書いたものです。