Round a subnormal product or quotient only once

[ibmibmibm]

• In the default mode, the divide of decimal32_t, decimal64_t and decimal128_t, and
  the multiply of decimal64_t, rounded the wide result to the precision of the type
  first. For a subnormal result, the constructor then rounded that value again to the
  smallest exponent. Some results were one unit away from the correct value.
• The four finalizers now compare the result exponent with etiny_v first. Below it, the
  divide gives the constructor the quotient with one more digit. If the remainder is not
  zero, this digit is 1. The multiply gives the exact product. The constructor then
  rounds once.
• The divide sends this value through its existing pack_in_range call. The fast types
  flush these results to zero, thus the guard is not in their code.
• The new test github_issue_1473.cpp checks 12 subnormal quotients and products. All 12
  fail before the change.

After the change, a sweep of subnormal products and quotients gives no wrong result in
the five rounding modes. The time of the multiply and the divide changes by less than 5
percent. Issue #1473 gives the values, the sweep, the comparison with develop and the
benchmark.

Fixes #1473

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.6%. Comparing base (8456280) to head (897162f).

Additional details and impacted files

@@            Coverage Diff            @@
##           develop   #1474     +/-   ##
=========================================
+ Coverage     98.6%   98.6%   +0.1%     
=========================================
  Files          307     308      +1     
  Lines        25703   25727     +24     
  Branches      2205    2205             
=========================================
+ Hits         25338   25364     +26     
+ Misses         365     363      -2     

Files with missing lines	Coverage Δ
include/boost/decimal/detail/div_impl.hpp	92.0% <100.0%> (+0.4%)	⬆️
include/boost/decimal/detail/mul_impl.hpp	95.2% <100.0%> (+0.1%)	⬆️
test/github_issue_1473.cpp	100.0% <100.0%> (ø)

... and 1 file with indirect coverage changes

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8456280...897162f. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[mborland]

Thanks as always!