Increase accuracy of parsing subnormal floats

[quinnj]

Fixes #83. Previously, we eagerly bailed for small enough exponents when
parsing. Instead, we can take a slightly slower path by falling back on
BigInt/BigFloat for however small the exponent is to do the correct
scaling.

[codecov]

Codecov Report

Merging #92 (2750564) into main (f1c4530) will decrease coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #92      +/-   ##
==========================================
- Coverage   87.56%   87.52%   -0.04%     
==========================================
  Files           9        9              
  Lines        2267     2269       +2     
==========================================
+ Hits         1985     1986       +1     
- Misses        282      283       +1     

Impacted Files	Coverage Δ
src/floats.jl	92.78% <100.00%> (-0.27%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f1c4530...2750564. Read the comment docs.