The logging is activated by default in Debug mode, and can be
activated by defining the preprocessor macro `HPP_FCL_ENABLE_LOGGING`.

This whole rewrite of GJKSolver is motivated by the fact that when using
an instance of GJKSolver multiple times, it mallocs data for EPA each time.
Note: if working with meshes, then a malloc also occurs in GJK (in the
support function).
This is not ideal for objects like `ComputeCollision` and `ComputeDistance`,
which are functors defined per single pair of collision, and which both
contain a `GJKSolver`.
(These functors are meant to be re-used so as their internal GJKSolver.)

This commit fixes this malloc issue by doing the following:
- It adds an instance of `GJK` and an instance of `EPA` in the
`GJKSolver` class.
- It leverages the newly introduced `reset` method of EPA to allocate
memory **only when needed**.
This last point is important because some functions of the `GJKSolver`
class have specializations which don't use EPA (and/or GJK), so we don't
want to allocate memory in these cases.

In the future, to lower the memory footprint of the `GJKSolver` class,
we could switch to a `data` model like in Pinocchio.
That way, `GJK` and `EPA` would only be thin interfaces implementing
the algorithms.
They would be working on a `data` object that allocated by the `GJKSolver`
when needed.

There is no need to use a `ShapeDistanceTraversalNode` for a collision
pair of two primitive shapes.
We do the same thing here for ShapeShapeDistance as we did for
`ShapeShapeCollide`.

Add option to turn asserts into exceptions and fix EPA related assert

[pre-commit.ci] pre-commit autoupdate

updates:
- [github.com/astral-sh/ruff-pre-commit: v0.2.0 → v0.2.1](https://github.com/astral-sh/ruff-pre-commit/compare/v0.2.0...v0.2.1)

Just like GJK, the result of EPA should always be checked against its
own tolerance.

Setting default parameters for the narrowphase

time

I only keep the cases that were previously triggering asserts.
Otherwise this test timeout on the CI.

All of the other cases don't trigger any assert.

Also bumped the tolerance of the assertion because asking 1e-16 is way
too small.

Checking the max coef is incompatible with the first convergence
criterion of GJK which checks the norm of the current iterate (see first
convergence check in `GJK::evaluate` in `gjk.cpp`).
This convergence check is needed, so it's better to adapt the assertion
than the GJK check itself.

Avoid double fwd declaration in Pinocchio eigen.hpp