increase signal pad size for dist matmul kernels

[shunting314]

Stacked PRs:

• Use a larger timeout setting to autotune distributed kernels #1800
• move compile time dist APIs to _dist_utils.py #1799
• properly codegen hl.triton_kernel #1797
• cleanup the autotuner factory function #1770
• remove hl.signal/wait #1791
• restrict to persistent pid for dist kernels #1772
• move distributed runtime utils out of examples/ folder #1771
• ->increase signal pad size for dist matmul kernels #1753

---

increase signal pad size for dist matmul kernels

the default signal pad size would cause out of bound global memory write if autotuner pick small block size. Increase the default for the matmul kernels. I'll improve autotuner to not autotuner beyond the signal pad size in a followup PR.