respect $IterationLimit more places

[rocky]

Fixes #1476

There are probably more places where we need to do something similar, but this is a start.

[mmatera]

In[4]:= FindMaximum[-Exp[-1/x^2]+1., {x,1.2}, MaxIterations->2]                 

FindMaximum::cvmit: 
   Failed to converge to the requested accuracy or precision within 2
     iterations.

Out[4]= {1., {x -> 0.168461}}

[rocky]

Ok. For FindMaximum, at iterations 2 it fails. At 10 though it doesn't which is what the test was.

Also, when it fails, it gives a value, and Mathics3 does not.

[rocky]

@mmatera: I am seeing here that we get a lot of Iteration limit occurrences in FindRoot, and one in combinatorica testing. I haven't had time to track these down.

But there is a bit here that should be done independent of improving $InterationLimit. As too often is the case, in looking at failures, there is a bit of other stuff that should be improved.

In particular, I think we should set the IterationLimit to 4096 as it is in WMA. And then to deal with the slowness of tests that were we to hit 4096, we set that in a Module with a local setting.

So at some point I'll split this PR in two.

[mmatera]

In MathicsSession, the same evaluation object is used in all the evaluations of the session. At which point self.iteration_count is reset to 0?

[rocky]

Ok - thanks. So we should reset this on each test.

We've added a number of other fields too, so those should probably be reset as well.

[mmatera]

In the old Mathics, one Evaluation object was created each time a line in the CLI or a cell in the web interface was executed. At some point, we change this behavior because most of the time, the Evaluation object does not change between evaluations. Now, since the evaluation always starts with a call to evaluation in the MathicsSession object, it is sufficient to reset the value there.

[rocky]

@mmatera if this is okay to go as one PR, great. If not, I'll split it in two tomorrow along the lines mentioned before.

[mmatera]

@mmatera if this is okay to go as one PR, great. If not, I'll split it in two tomorrow along the lines mentioned before.

I think this is OK for now. Then, I can try to fix the output of FindMinimum and friends.

[mmatera]

LGTM. @rocky, merge this when you feel it is ready.