Fix several printing segfaults.

[alexreinking]

Fixes #8699

[alexreinking]

It would be good to audit all our Python __repr__ and __str__ functions to check that they (a) don't crash (at a minimum) and (b) behave like Pythonic __repr__ and __str__ functions... i.e., __repr__ should be something copy-paste-able into a REPL (maybe with from halide import *)

[alexreinking]

An LLM success story: I asked Claude Code to audit all the __str__, __repr__, and operator<< functions in the codebase for .defined()-ness issues. It turned up a few false-positives, but it found the Pipeline issue and the exercise helped me notice and construct the Func::name() segfault.

Who knows if it found them all?

No code in this PR was written by Claude; I asked it only to find fishy functions.

[mcourteaux]

This is to prevent updating Halide.h when it didn't change, and don't recompile all dependents? I like :)

[alexreinking]

This is to prevent updating Halide.h when it didn't change, and don't recompile all dependents? I like :)

It also seemed to fix an issue on the arm mac worker

[alexreinking]

Lots of unrelated failures:

• Make failing due to a new layout for Homebrew's libpng & libjpeg.
• Make failing due to -Werror in the previously untested onnx app (denormal_disabler.h). Should be fixed by Add CMake for onnx app #8707.
• LLVM trunk simd-op-check failures
• On arm-32-linux, a failure in correctness_mul_div_mod appears to be new
• adams2019_test_autoscheduler performance test is known to be flaky.
• LLVM 21 has an error in correctness_register_shuffle
• WASM has an issue with correctness_indexing_access_undef

Lots to follow up on, but nothing to halt overall progress.