Implement rounding functions for SciFloat

[matterhorn103]

These are required for the SciNum implementation. They make less sense for SciFloat than for SciDecimal since f64 doesn't properly maintain precision, but a best effort can still be made.

• precision()
• precision_most_significant_fig()
• precision_uncertainty()
• dp()
• sf()
• sf_uncertainty()
• round_precision()
• round_dp()
• round_sf()
• round_match_uncertainty()
• round_match_uncertainty_sf()
• round_uncertainty_precision()
• round_uncertainty_dp()
• round_uncertainty_sf()
• round_uncertainty_match_number()
• trunc_sf()