Precise flonum tests (error usually must be <= 0.5 ulp), and prerequisite
additions/changes
* More accurate `flulp-error'
* Added `flonum->fields', `fields->flonum', `flonum->sig+exp',
`sig+exp->flonum' (currently undocumented)
* Correctly rounded, robust `bigfloat->fl2' and `fl2'
* Correctly rounded, robust `fl+/error', `fl-/error', `fl*/error',
`flsqr/error', `fl//error'
* Much faster but slightly less accurate fl2 ops (shamelessly stolen from
crlibm, which is LGPL)
* Added `fl2ulp', `fl2ulp-error', `fl2?' (which detects overlap),
`+max-fl2-subnormal.0' (which was tricky), `fl2abs'
* Added deterministic and randomized flonum op tests (against MPFR)
* Added deterministic and randomized flonum/error op tests (against MPFR)
* Added deterministic and randomized fl2 op tests (against MPFR)
* Exposed FPU tests in `math/utils' (currently undocumented)
e55a31480e