fix: prevent arm64 OOM during linking with lld + job limit

[zeekay]

Summary

• Configure lld as the linker for Linux targets in .cargo/config.toml — lld uses ~3x less memory than the default ld linker, which is critical for the arm64 Graviton runner (c7g.2xlarge, 16GB RAM)
• Add clang to apt-get install (required as the linker driver for -fuse-ld=lld)
• Set CARGO_BUILD_JOBS=4 in both Dockerfiles to cap parallel compilation and further reduce peak memory

Root cause

The arm64 Docker build OOMs during the final link step (collect2: fatal error: ld terminated with signal 9 [Killed]). The default GNU ld linker loads all object files into memory at once and on arm64 with hundreds of crates this exceeds 16GB. The amd64 build succeeds because the DOKS runner pod has more available memory.

Changes

File	Change
.cargo/config.toml	Add [target.x86_64-unknown-linux-gnu] and [target.aarch64-unknown-linux-gnu] sections using clang+lld
docker/Dockerfile	Add clang to build deps, add ENV CARGO_BUILD_JOBS=4
.github/Dockerfile	Add clang to build deps, add ENV CARGO_BUILD_JOBS=4

Test plan

• arm64 Docker build completes without OOM
• amd64 Docker build still succeeds (lld is faster there too)
• CI test image builds on both architectures