Fix random seeding

[mikeyford]

See monome/teletype#337

The use of a constant to initialize z meant the first random values aren't sufficiently different for similar seeds. I'm not very handy with C, so a Python script was easier to demonstrate:

class RandomState:
    def __init__(self, seed):
        self.z = 12345  
        self.w = seed  

    def random_next(self):
        self.z = 36969 * (self.z & 0xFFFF) + (self.z >> 16)
        self.w = 18000 * (self.w & 0xFFFF) + (self.w >> 16)
        return ((self.z << 16) + self.w) & 0x7FFFFFFF

for seed in range(10):
    rand = RandomState(seed)
    print(rand.random_next())

returns

1465974784
1465992784
1466010784
1466028784
1466046784
1466064784
1466082784
1466100784
1466118784
1466136784

The values aren't exactly the same, but are similar. This effect disappears on further iterations of the initial seed. In specific situations like PN.RND where we take the mod of the random value this manifests as the same value each time on the first call. For ops like RAND 10000 the initial values were different on reseeding, but tended to be clustered.

Updating to the change in this PR:

class RandomState:
    def __init__(self, seed):
        self.z = ~seed  
        self.w = seed  

and re-running we get the values

1872101376
1596802640
1321503904
1046205168
770906432
495607696
220308960
2092493872
1817195136
1541896400

Built firmware and tested in Teletype hardware.

[tehn]

Thanks for spotting this.