Add support for FixedPointParameters with negative number of integral bits.

[ghost]

This is required to support fixed point formats with value ranges like [-0.5, 0.5), e.g.
Q-1.16 on 16 bits:
0x8000 -> -0.5 (32768/2^17)
0x7FFF -> 0.499984

Q-2.17 on 16 bits:
0x8000 -> -0.25 (32768/2^17)
0x7FFF -> 0.249992

For those formats number of integer bits needs to be set to negative value.

[sguiriec]

Can you clarify why you have put these numbers of 16 bits container and not 32 bits?

Q format is defined as per https://en.wikipedia.org/wiki/Q_%28number_format%29

So Q1.16 number should be define on 32 bits container.

[ghost]

"For a given Qm.n format, using an m+n bit signed integer container with n fractional bits..."
m can also be negative if one want to represents small numbers. In my examples I did not include the sign bit in the m value, after adjusting to match the wikipedia samples it would be:

1. Q0.16: m = 0; n = 16 ; container m+n (0 + 16) -> 16bit; range [-(2^(0-1), 2^(0-1)- 2^-16]; resolution 2^-16

2. Q-1.17: m = -1; n = 17 ; container m+n (-1 + 17) -> 16bit; range [-(2^(0-2), 2^(0-2)- 2^-17]; resolution 2^-17

Similar examples can be created for 32 bit containers. The issue is about support for fixed point parameters where the resolution is being increased by limiting the range of representable values.
For example, using 16bit containers for brevity:

Q16.0, range [-32768, 32767), resolution 1  (degenerative case)
...
Q2.14,  range [   -2,    2), resolution 2^-14
Q1.15,  range [   -1,    1), resolution 2^-15
Q0.16,  range [-0.5, 0.5), resolution 2^-16
Q-1.17, range [-0.25, 0.25), resolution 2^-17
Q-2.18, range [-0.125, 0.125), resolution 2^-18
...