Test corner cases of parser

[Flamefire]

Inspecting the function I see some corner cases that could yield wrong values and/or overflow.

What I suspect:

• There is an off-by-one error in the buffer size which causes early truncation of the significant
• The exponent might overflow if it is too large. By using a type with at least 20bits this likely does not happen in practice, so I guess that should be enforced.
• However for large strings the exponent might still overflow or even e.g. leading_zero_powers or extra_zeroes might already

I came up with some tests that should exercise the corner cases. Please check if they make sense so far before I go into more details