Avoid Importing Fastmath Functions

[seanlaw]

Currently, when we set fastmath default flags in __init__.py:

stumpy/stumpy/__init__.py

Lines 41 to 48 in 97da018

	if not numba.config.DISABLE_JIT: # pragma: no cover
	njit_funcs = cache.get_njit_funcs()
	for module_name, func_name in njit_funcs:
	module = importlib.import_module(f".{module_name}", package="stumpy")
	func = getattr(module, func_name)
	key = module_name + "." + func_name # e.g., core._mass
	key = "STUMPY_FASTMATH_" + key.upper() # e.g., STUMPY_FASTHMATH_CORE._MASS
	config._STUMPY_DEFAULTS[key] = func.targetoptions["fastmath"]

We are actually importing every single njit function. Instead, we should use AST to read the decorator value and then cross reference the values that they are represented by in config.py.