Don't use *ll* and *ul* in whole-intsubs
`whole-intsubs` was calling `make-defint-assumptions` with `*ll*` and
`*ul*`. However the args `a`, and `b` to `whole-intsubs` are the
limits of the definite integral, so pass `a` and `b` to
`make-defint-assumptions`.
I manually looked through the call tree for `whole-intsubs` and
AFAICT, `a` and `b` are always assigned to the value of `*ll*` and
`*ul*` eventually. However, there are some cases where `a` is not
`*ll*` and `b` is not `*ul*`. I don't understand why.
With this change, the full testsuite + share passes with cmucl and
ccl64 except for rtestint problem 237:
```
********************** Problem 237 (line 1428) ***************
Input:
integrate(exp(sqrt(x) - x), x, 0, inf)
Result:
1/4 1 1
%e gamma_incomplete(─, ─)
1/4 1 2 4 1/4
%e gamma_incomplete(1, ─) - ──────────────────────────── + %e sqrt(%pi)
4 2
This differed from the expected result:
1/4 1 1
%e gamma_incomplete(─, ─)
1/4 1 2 4
%e gamma_incomplete(1, ─) - ────────────────────────────
4 2
1/4 1/4
%e (sqrt(%pi) + 2) %e sqrt(%pi) 1/4
+ ───────────────────── + ─────────────── - %e
2 2
```
If you factor the expected result, you get the actual result. Plus if
you look at rtestint, there's a comment that says the actual result is
equivalent to what we have. That's nice sign that the change to
`whole-intsubs` is ok, despite the `a` and `b` not being the same as
`*ll*` and `*ul*` in some cases.
I didn't update the expected result yet.