Revert "[codegen] annotate member-access expressions with class/interface types"

[cs01]

Before

#662 merged with all PR checks green on arm64 (local + macOS CI). After merge to main, CI on x86-64 Linux + macOS segfaults at Stage 0 → Stage 1 self-hosting:

```
Command failed: /tmp/chad-stage0 build -v src/chad-native.ts -o /tmp/chad-stage1 --target-cpu=x86-64
Segmentation fault (core dumped)
```

The member_access annotation generates IR that works on arm64 but segfaults the stage-0 compiler on x86-64. Likely an undefined-behavior case that `-O2` + x86-64 target exploits differently than arm64.

After

Main passes CI again. Gate-loosening step 2 (`member_access`) reverted; will reland after x86-64 repro.

Description

PR #662 slipped past its own CI because the PR-level matrix runs macOS+arm64 only (per my observation — `build-macos pass`, `build-linux-glibc pass` on the PR, but then `build-linux-glibc fail` after merge on main CI). Need to either add x86-64 Linux to PR CI or test locally with `--target-cpu=x86-64` before merge.

Refs #658.

[github-actions]

Benchmark Results (Linux x86-64)

Benchmark	C	ChadScript	Go	Node	Place
Fibonacci	0.816s	0.765s	1.562s	3.176s	🥇
Hash Map Lookup	0.091s	0.063s	0.090s	0.112s	🥇
Binary Trees	1.576s	1.322s	2.804s	1.207s	🥈
Cold Start	0.8ms	0.9ms	1.3ms	28.2ms	🥈
File I/O	0.118s	0.092s	0.086s	0.209s	🥈
JSON Parse/Stringify	0.038s	0.053s	0.177s	0.135s	🥈
N-Body Simulation	1.711s	2.138s	2.215s	2.390s	🥈
Regex Match	0.016s	0.005s	0.021s	0.004s	🥈
SQLite	0.052s	0.368s	0.484s	0.422s	🥈
Monte Carlo Pi	0.390s	0.411s	0.406s	2.253s	🥉
Quicksort	0.216s	0.247s	0.213s	0.263s	🥉
Sieve of Eratosthenes	0.016s	0.027s	0.020s	0.038s	🥉
String Manipulation	0.008s	0.019s	0.017s	0.039s	🥉
Matrix Multiply	0.450s	0.991s	0.640s	0.375s	#4

CLI Tool Benchmarks

Benchmark	ChadScript	grep	node	xxd	Place
Hex Dump	0.652s	—	1.058s	0.132s	🥈
Recursive Grep	0.023s	0.011s	0.108s	—	🥈