avoid exceeding numfg in one step

[XingyuZhang2018]

see the issue in #26

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.35%. Comparing base (b3ba9f8) to head (12becab).

Additional details and impacted files

@@           Coverage Diff           @@
##           master      #27   +/-   ##
=======================================
  Coverage   82.35%   82.35%           
=======================================
  Files           5        5           
  Lines         527      527           
=======================================
  Hits          434      434           
  Misses         93       93           

☔ 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.

[Jutho]

I think this looks good. The reason that I didn't implement this earlier, is because the bisect and braket routines can be called at any point during the linesearch. Their local value of numfg is not the total number of function/gradient evaluations along the linesearch. So even with the current check, will still allow for the linesearch to exceed maxfg.

However, I fully agree that the current check will already be much better and make the exceeding of maxfg much smaller, so I will happily merge this. A full solution to make the linesearch obey the maxfg limit strictly will require to change the implementation logic of the linesearch.

So I will happily merge this PR. My apologies for the delay.