Poor random numbers in uint tests.

[jsallay]

https://github.com/gnuradio/volk/blob/main/lib/qa_utils.cc#L54 Contains: float int_max = float(uint64_t(2) << (type.size * 8));. If type.size is 8, then int_max = uint64_t(2) << 64. Shift a 64-bit number by 64 is undefined behavior. It appears that gcc masks the shift amount and no shift is performed. Thus the random numbers generated are between 0 and 2 for unsigned 64-bit numbers and -1 and 1 for signed 64-bit numbers.

This is not ideal and causes some tests to not actually exercise algorithm. See #605

[jdemel]

This line is part of:

volk/lib/qa_utils.cc

Line 41 in e853e9b

void load_random_data(void* data, volk_type_t type, unsigned int n)

My idea to fix this issue would be:

1. Implement another function for every type
2. Add an optional seed for reproducible results
3. Call the specialized functions from load_random_data

The structure is created here:

volk/lib/qa_utils.cc

Line 123 in e853e9b

volk_type_t volk_type_from_string(std::string name)

And this is

volk/lib/qa_utils.h

Lines 25 to 32 in e853e9b

	struct volk_type_t {
	bool is_float;
	bool is_scalar;
	bool is_signed;
	bool is_complex;
	int size;
	std::string str;
	};

Some templating may reduce the number of specific functions.

This issue is related to #326 .