Fix WorkSpace issues with expression optimization

[jodavies]

Here are two changes to make the optimizer easier to use.

The first is to dynamically allocate memory in Horner_tree, if the WorkSpace does not have enough space. There is no particular reason that the workspace should be used for the sort, here, other than that it exists, and might have capacity.

The second, is to re-use the same WorkPointer for every call of Generator when the optimized expression is placed into FORM's output. Generator modifies AT.WorkPointer internally, so each term which was added here was placed further and further into the buffer, unnecessarily.

In principle, if you arrive in generate_expression with a very-nearly full WorkSpace there could still be a crash. But this should be rare (?) .

These changes pass pySecDec's test suite.

[coveralls]

coverage: 49.977% (-0.02%) from 49.999%
when pulling 603737d on jodavies:opt-dynamic-workspace
into 83e3d41 on vermaseren:master.

[tueda]

For the first commit, we can add a test, for example:

#:maxtermsize 200
#:workspace 50000
S x1,...,x19;
L F = (x1+...+x19)^4;
Format O1;
.sort
#optimize F
L G = `optimvalue_';
P G;
.end
assert succeeded?
assert result("G") =~ expr("389")

which fails for the master branch.

I am not 100% sure we can reuse the workspace in the second commit, but the absence of test failures suggests that it is working.

[tueda]

The last commit (for check/fixes.frm) has a conflict but should be resolved easily.

[coveralls]

coverage: 50.199% (+0.09%) from 50.11%
when pulling b4925e2 on jodavies:opt-dynamic-workspace
into 3c5b5ce on vermaseren:master.