remove require_cuda defaults in internal functions

[jameslamb]

Contributes to rapidsai/build-planning#31.

This project attempts to determine the version of CUDA present on the system using nvcc. By default, if that attempt fails it raises an exception.

This is configurable from the public API of the build backend, like this in pyproject.toml...

[tool.rapids-build-backend]
require-cuda = true

rapids-build-backend/README.md

Line 25 in 0c7e6d1

| `require-cuda` | If false, builds will succeed even if nvcc is not available | bool | true | Y |

This PR proposes removing defaults in the signatures of internal functions that perform implement that behavior, to reduce the risk of "oops forgot to pass this through and rapids-build-backend silently fell back to some a default" types of bugs.

In my opinion, it'd be useful to restrict knowledge of default values for configuration to exactly one place... the code used to parse configuration.

rapids-build-backend/rapids_build_backend/config.py

Lines 26 to 37 in 0c7e6d1

	# Mapping from config option to default value (None indicates that option is
	# required) and whether it may be overridden by an environment variable or a config
	# setting.
	config_options: "config_options_type" = {
	"build-backend": (None, False),
	"commit-file": ("", False),
	"dependencies-file": ("dependencies.yaml", True),
	"disable-cuda-suffix": (False, True),
	"matrix-entry": ("", True),
	"require-cuda": (True, True),
	"requires": (lambda: [], False),
	}

And in general, for there to be as few competing sources of configuration values as possible.

How I tested this

Looked for all uses like this.

_get_cuda

Otherwise, relying on this project's pre-commit + CI.