ENH: integrate.cubature: support for infinite limits (scipy#21657)

* ENH: add `points` for subdividing initial region

- Add `points` argument to `cubature`
- Break up the region used for the initial estimate so that all `points`
  lie on the edge of a subregion
- Add tests using the `points` argument

* ENH: support infinite limits via variable transformation

- Handle limits where a > b by flipping them around and keeping track of
  the sign
- Add `_VariableTransform` base class
- Add `_InfiniteLimitsTransform` class
- Apply infinite limits transformation if any of `a` or `b` are infinite
- Add test for limits where a > b
- Add test for infinite limits
- Add test to check that break points are correctly mapped under
  infinite limits transformation

* DOC: add details of `points` and infinite limits to docstring

- Update docstring to say that infinite limits are allowed in `a` and
  `b`
- Add `points` argument to Parameters section
- Describe the transformation used to handle infinite limits in the
  docstring
- Add example of 2D integral with infinite limits
- Add example of 1D integral with singularities avoided using `points`

* TST: call `cubature` using keyword arguments in new tests

* DOC: add more detail to `points` parameter

* DOC: rephrase to "Jacobian determinant" rather than "Jacobian"

* MAINT: pass `xp` as argument rather than calling `array_namespace`

* MAINT: return results in the common dtype of input

- Store nodes and weights by default using `float64`
- Downcast if necessary to match the common floating dtype of user input
- Add tests for this

* MAINT: pass `xp` to `_apply_fixed_rule` rather than using `array_namespace`

* STY: `len(...) == 0` rather than comparing to `[]`

* TST: tighten tolerance on some tests

Tighten tolerances from 1e-1 to 1e-8 on the following tests:

- `test_pass_str`
- `test_pass_array_like_not_array`
- `test_zero_width_limits`
- `test_limits_other_way_around`
- `test_base_1d_quadratures_simple`
- `test_base_1d_quadratures_error_from_difference`

* DOC: correct details of transformation used for infinite limits

- Correct details of transformation used in the (\infty, b_i] case
- Remove details of Jacobian determinant since it is not true when the
  limits are (\infty, b_i]

* STY: remove extra blank line

* MAINT: remove unnecessary `negate_pos` in transformation

- `negate_pos` wasn't having any effect since the values that were
  negated would be overriden below

* DOC: correct details for (-\infty, b_i]

* remove for loop from __init__

* still bugs to figure out

* Update scipy/integrate/_cubature.py

Co-authored-by: Olly Britton <ollybritton@gmail.com>

* fix axis location

* fix summing boleans

* don't use arr[..., bool_array]

* BUG: use compatible namespace in test for infinite limits and points

* BUG: correctly map break points when using a transformation

- Need to reshape points so that they can be mapped by `f.inv`
- Add test case for 1D and 2D integrals where the points specified are
  removed by the transformation
- Adjust double infinite transformation to prevent division by zero

* MAINT: explain check for any infinite limits in `points`

[lint only]

---------

Co-authored-by: Jake Bowhay <jb9.bowhay@gmail.com>