feat: testbench core infrastructure (PLENAProgram, DeveloperCompiler, SubMatrixManager)

[booth-algo]

Summary

3-layer compilation stack for generating PLENA ISA from high-level tensor operations:
(Developed by @gaoziqian123)

• PLENAProgram (1.5k lines): tensor proxy API, HBM auto-allocation, scoped naming, @ operator overloading
• DeveloperCompiler (2.7k lines): register allocation, ISA string emission, ASM template invocation
• SubMatrixManager (2k lines): VRAM/MRAM/HBM memory layout, sub-block addressing, virtual memory allocators

Supporting infrastructure:

• symbol_table.py — tensor metadata tracking
• config_utils.py — plena_settings.toml read/write
• emulator_runner.py — build + run emulator, compare against golden
• check_mem.py — VRAM/golden comparison with allclose metrics
• Rust emulator: new opcodes (V_SHFT_V, H_STORE_V_PART)
• justfile: test recipes for all operator tests
• CLAUDE.md: project context + CI check instructions

Architecture

PLENAProgram (tensor proxy API, HBM allocation)
  → DeveloperCompiler (register allocation, ISA emission)
    → SubMatrixManager (VRAM/MRAM/HBM memory layout)
      → ISA code string

Test plan

• Infrastructure only — no standalone tests (tests come in PR feat: ATen-style ops dispatch, compiler, and full test suite #5: kev/aten-ops)
• Code review of compilation stack

⚠️ nix-build CI fails due to pre-existing DNS issue. Unrelated to this PR.