Special values as generator indices can cause zero generators

[mbudisic]

Even though we disallowed "0" as input to braid, zero generators can still apparently be created.

Also inf can safely be passed to braid as well. We should consider disallowing special values as well or, alternatively, skipping them without error. For debugging purposes, I think it would be better if NaN and Inf were treated as errors in input.

Currently the situation looks like this:

>> braidlab.braid([1, nan, 2])
ans = 
 < 1  0  2 >
>> braidlab.braid([1, inf, 2])
ans = 
 < 1  2147483647           2 >
>> braidlab.braid([1, -inf, 2])
ans = 
 < 1 -2147483648           2 >
>> braidlab.braid([1, -nan, 2])
ans = 
 < 1  0  2 >

[jeanluct]

This is due directly to

>> int32(inf)

ans =

  2147483647

int32, unlike double, cannot have a representation for infinity. Also

>> int32(nan)

ans =

           0

That one's a bit weird! Again, int32 has no "redundant" codes that can be used to hide special meaning, as for double. I'll fix this.