fix precision of non-integer power

[araujoms]

When raising an integer to DoubleFloat power precision was lost, because the log was computed as a Float64. This PR fixes this.

Before the PR:

julia> 2^Double64(0.5) - sqrt(big(2))
-1.639813727959775997377421317343763431286407155713134856781850122263015902218541e-17

After the PR:

julia> 2^Double64(0.5) - sqrt(big(2))
-4.138675308699413612074474846792288426668019534744073764445159044568402999352625e-33

A bit unrelated, but why are you converting to Float64 before converting to Int64 here?

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 49.17%. Comparing base (2ef0305) to head (95c89e1).
⚠️ Report is 6 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #227      +/-   ##
==========================================
- Coverage   49.20%   49.17%   -0.04%     
==========================================
  Files          63       63              
  Lines        3410     3406       -4     
==========================================
- Hits         1678     1675       -3     
+ Misses       1732     1731       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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

[JeffreySarnoff]

thank you.
If you could, be very specific (file, lines) about
"A bit unrelated, but why are you converting to Float64 before converting to Int64 here?"

[araujoms]

It was in the line 124 of the file src/math/elementary/explog.jl:

   return r^Int64(Float64(n)) # convert n to Float64 first

it is not there anymore because I deleted it, as it became unnecessary due to the cutoff I introduced in the size of the exponent. Now I just take n.hi.