[ENH] Fix printing values with too few decimals

[janezd]

Issue

Fixes #4555.

Description of changes

Print two more decimals when the format is not %g and the difference from the rounded values is outside numeric error.

Includes

• Code changes
• Tests

[codecov]

Codecov Report

Merging #4575 into master will decrease coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #4575      +/-   ##
==========================================
- Coverage   83.59%   83.56%   -0.03%     
==========================================
  Files         281      276       -5     
  Lines       56131    55331     -800     
==========================================
- Hits        46921    46236     -685     
+ Misses       9210     9095     -115     

[markotoplak]

This PR makes printing tables slower, but this operation does not need to be fast. It does not influence saving data, because the writer has its own formatter. Here are the differences between master and this branch:

[str_val_decimals] with 10 loops, best of 3:
	min 797 usec per loop
	avg 824 usec per loop
[str_val_g] with 10 loops, best of 3:
	min 681 usec per loop
	avg 728 usec per loop
[print_dense] with 10 loops, best of 3:
	min 1.98 msec per loop
	avg 2.02 msec per loop
[print_dense_decimals] with 10 loops, best of 3:
	min 2.1 msec per loop
	avg 2.16 msec per loop


[str_val_decimals] with 10 loops, best of 3:
	min 2.51 msec per loop
	avg 2.7 msec per loop
[str_val_g] with 10 loops, best of 3:
	min 1.01 msec per loop
	avg 1.04 msec per loop
[print_dense] with 10 loops, best of 3:
	min 2.45 msec per loop
	avg 2.47 msec per loop
[print_dense_decimals] with 10 loops, best of 3:
	min 11.7 msec per loop
	avg 11.9 msec per loop

[janezd]

Excellent, thanks!

And even these differences are not too large.