Move tests to tests/, fix BINARY_OP for 3.12+

[sjbaebae]

Summary

• Move test files into tests/ with pyproject.toml pytest config
• Fix strict tracer missing BINARY_OP handler (Python 3.12+ unified opcode)
• Add TODO.md documenting known CPython 3.12 f_trace_opcodes bug and sys.monitoring migration path

Test results

• 32 pass, 1 known failure on Python 3.12 (test_my_add_basic — CPython bug Behavior change for opcode trace after PEP 669 python/cpython#103615)
• All pass on Python 3.11 and 3.13+

Test plan

• pytest tests/ -v on Python 3.12
• Manual test run on Python 3.11 (all 33 pass)

[SethTS]

Seth's ClaudeCode reviewed the PR and had a suggestion:

Nice fix — the BINARY_OP addition makes sense and is clearly the right call.

One question: would it be worth just requiring Python 3.13+ and dropping 3.11/3.12 support entirely?

The reasoning: 3.12 has the known f_trace_opcodes bug (CPython #103615) that causes test_my_add_basic to fail, and 3.11 uses the old individual opcodes that this PR is partly working around. If we set requires-python = ">=3.13" in pyproject.toml, the known failure goes away, the test matrix collapses to a single version, and the sys.monitoring migration in your TODO becomes a nice-to-have rather than a necessary fix for 3.12 correctness.

The BINARY_OP fix in this PR is still needed on 3.13 (since 3.13 also uses the unified opcode), so this isn't saying the fix is wrong — just asking whether we could simplify the supported range at the same time.

Our env is sutro313 (Python 3.13.5) and the competition pipeline would presumably run on whatever Python we specify in CI, so there's no strong reason to carry the older versions if they're causing known failures.