[FIX] Let the tile-* CLIs run without pytest installed

[mark14wu]

Summary

• tilelens.wrapper imported pytest at module level, but pytest is only in the test extra. After a plain pip install tilelens, tile-sanitizer, tile-profiler, tile-race and their triton-* aliases all failed with ModuleNotFoundError: No module named 'pytest' before running anything. This was already the case for triton-viz 3.0, and it is visible now that tilelens 3.1 is on PyPI.
• Import pytest only in the branch that forwards to pytest.main, the one place it is used.
• Bump the version to 3.1.1 so the fix can be released.

Test Plan

• Added test_wrapper_imports_without_pytest, which imports tilelens.wrapper in a subprocess with pytest blocked. The same check fails on main.
• Built the 3.1.1 wheel and installed it with torch and numpy but without pytest. tile-sanitizer, tile-profiler, tile-race, triton-sanitizer, triton-profiler, and triton-race-detector all ran a small kernel script.
• After installing pytest, tile-sanitizer pytest -q test_k.py still runs the test.

After merge: tag the merge commit v3.1.1 and publish a GitHub release, which uploads tilelens 3.1.1 through the existing trusted publisher.

Related Issues

Follow-up to #477.

Breaking Changes

None.

Checklist

• I added tests to all new functionality I added/bugs I fixed.
• I verified that a human has reviewed all code in this PR.
• I ran npm run build:frontend if the PR modified any TypeScript code.
• I made sure that my code is well documented (comments explaining strange code, docstrings for functions, website modified if new functionality added).

[github-actions]

Performance Benchmark

Benchmark	main (min)	PR (min)	Change	Samples
gemm	0.106s	0.105s	-0.6%	20 / 20
gemm_oob	0.118s	0.118s	-0.3%	20 / 20
indirect_load	0.022s	0.022s	-0.1%	20 / 20
nested_loop	0.237s	0.236s	-0.4%	20 / 20
block_pointer_loop_advance	0.126s	0.126s	-0.3%	20 / 20
liger_jsd	0.140s	0.140s	-0.1%	20 / 20
flaggems_layernorm	0.396s	0.402s	+1.4%	20 / 20
swiglu	0.172s	0.172s	+0.4%	20 / 20
cross_entropy	0.991s	0.994s	+0.3%	20 / 20
fused_linear_jsd	0.211s	0.212s	+0.3%	20 / 20
Total	2.520s	2.527s	+0.3%	N/A

Iterations: 1 warmup + 20 measured
Samples are shown as main / PR; long pytest benchmarks may use fewer samples.