Implement IRIS-NP2

[cohnt]

This adds the minimal functionality for IRIS-NP2 to begin being used. Additional features will be added following this PR, but we're already over 1000 lines. This does not close #21822.

---

This change is 

[cohnt]

+@RussTedrake for feature review

[cohnt]

@drake-jenkins-bot mac-arm-sonoma-clang-bazel-experimental-release, please

[RussTedrake]

A few small questions/requests below.
The code is beautiful. It's hard to check all of the details/logic, but I did try to do a pretty careful read through.
thanks for doing this!

Reviewed 11 of 11 files at r1, all commit messages.
Reviewable status: 7 unresolved discussions, needs at least two assigned reviewers, labeled "do not merge", commits need curation (https://drake.mit.edu/reviewable.html#curated-commits) (waiting on @cohnt)

---

planning/iris/iris_np2.h line 60 at r1 (raw file):

volume-in-collision is larger than ε

Pr[λ(P\Cfree)/λ(P) > ε] ⋞ δ.

I assume that your use of ⋞ is intentional, but it's atypical. Does \le really not work?

---

planning/iris/iris_np2.h line 88 at r1 (raw file):

solving this function. See AcquireLicense for more details.

This feature is still in development, so certain features are not yet supported.

nit: you're using features twice in two different ways in the same sentence.

---

planning/iris/iris_common.h line 236 at r1 (raw file):

namespace internal {

// See Definition 1 in the paper.

this is in iris_common.h. I don't think it's clear which paper you're referring to here.

---

planning/iris/test/iris_np2_test.cc line 22 at r1 (raw file):

using symbolic::Variable;

// Reproduced from the IrisInConfigurationSpace unit tests.

is there a plan (perhaps a TODO) to share more of the code between the test files?

---

planning/iris/iris_np2.cc line 51 at r1 (raw file):

  for (const auto& pair : sorted_pairs) {
    auto query_object = plant.get_geometry_query_input_port()
                            .template Eval<QueryObject<double>>(context);

Hopefully the caching makes this actually ok, but I think it's more clear/clean if it's outside of the for loop?

---

planning/iris/iris_np2.cc line 66 at r1 (raw file):

}

/* Check if any unsuppoorted features have been used, as well as any other

nit: typo: unsupported

---

planning/iris/iris_np2.cc line 205 at r1 (raw file):

    DRAKE_DEMAND(body_B_ptr != nullptr);

    if (!checker.IsCollisionFilteredBetween(*body_A_ptr, *body_B_ptr)) {

i don't understand something here. i thought GetCollisionCandidates should not return any pairs that are already filtered?
so I don't understand the point of this loop (and the difference between pairs and possible_pairs?
I assume there is something I've missed, but then perhaps you need to explain it to me (and future readers) by documenting it around here?