Round a value below the smallest subnormal in each rounding mode

[ibmibmibm]

• The constructors of decimal32_t, decimal64_t and decimal128_t had a special branch for
  a significand of one digit below the smallest exponent. It rounded a tie up, and it ignored
  the upward and the downward modes below the first step. The branch is removed. The general
  branch after it rounds these values correctly in each mode.
• If all digits of the coefficient drop, coefficient_rounding set it to zero, and it did not
  round. It now rounds a zero with a sticky bit through fenv_round. A directed mode can then
  give the smallest subnormal, and a zero input still gives zero.
• decimal128_t called coefficient_rounding only for a coefficient type of 128 bits or more.
  A narrower coefficient far below the smallest exponent read past the end of the pow10
  table, and decimal128_t{10, -6231} stopped the program with SIGFPE. It now takes the same
  path.
• For a 128-bit coefficient with few digits, coefficient_rounding divided in the narrow
  significand type. A large shift did not fit in that type. The quotient was wrong, or the
  divide by zero stopped the program, for example in a conversion from decimal128_t to
  decimal32_t. The narrow divide now also needs a shift that fits.
• The new test github_issue_1476.cpp checks the four cases for the three types, in the
  modes with a defect. Before the change it stops with SIGFPE.

A sweep checks values below the smallest exponent through the constructor, the string parser
and the multiply. After the change, it gives no wrong result in the five rounding modes. Issue #1476
gives the values, the sweeps, the comparison with develop and the benchmark.

Fixes #1476

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.7%. Comparing base (5b03306) to head (4d43e0f).

Additional details and impacted files

@@            Coverage Diff            @@
##           develop   #1477     +/-   ##
=========================================
+ Coverage     98.6%   98.7%   +0.1%     
=========================================
  Files          312     313      +1     
  Lines        26265   26251     -14     
  Branches      2264    2249     -15     
=========================================
+ Hits         25892   25896      +4     
+ Misses         373     355     -18     

Files with missing lines	Coverage Δ
include/boost/decimal/decimal128_t.hpp	97.2% <100.0%> (+1.0%)	⬆️
include/boost/decimal/decimal32_t.hpp	97.4% <100.0%> (+0.9%)	⬆️
include/boost/decimal/decimal64_t.hpp	97.8% <100.0%> (+1.0%)	⬆️
include/boost/decimal/detail/fenv_rounding.hpp	97.4% <100.0%> (+0.1%)	⬆️
test/github_issue_1476.cpp	100.0% <100.0%> (ø)

---

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 5b03306...4d43e0f. Read the comment docs.

🚀 New features to boost your workflow:

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

[mborland]

In the test file you only have testing for 3 of the 5 rounding modes. Can you please add paths so that all are hit? Thanks

[mborland]

In the test file you only have testing for 3 of the 5 rounding modes. Can you please add paths so that all are hit? Thanks

[ibmibmibm]

All rounding mode tests were added.

[mborland]

Looks good! Thanks!