PyTorch Inductor and precompile proposals

Two modest pull requests landed in the PyTorch repository on the Inductor and precompile paths. Both remain early proposals with only automated review traffic so far.

Always in-place user Triton kernels for Inductor

A new pull request in pytorch/pytorch proposes that user-defined Triton kernels always run in-place under Inductor. The change targets how custom kernels are scheduled and memory-handled during compilation. Developers maintaining Triton extensions or studying Inductor kernel fusion will want to track whether the always-in-place rule becomes the default.

DynamoTracer added to precompile capture API

Another pull request introduces a DynamoTracer and exposes its captures through the existing capture() entry point in the precompile layer. The addition is described as a small extension to the precompile API surface. Anyone wiring Dynamo-based tracing into ahead-of-time or cached compile flows should note the new tracer type.