fix(ccd): honor filter_contact_pair PhysicsHook during CCD (closes #754)

[nickyvanurk]

What

Makes Continuous Collision Detection consult the user's PhysicsHooks::filter_contact_pair hook, matching narrow-phase semantics. Previously, CCD would clamp a fast-moving body's motion at a predicted impact with a pair the user had filtered out, leading to the bug described in #754.

Why

Per the issue, "filtering in rapier is done in narrow phase only, for contacts and intersections (bodies and areas). The problem is in case of CCD we don't do any of these filterings." When a user enables CCD on an object and also installs a filter_contact_pair hook to suppress certain collisions, the hook is bypassed during CCD's time-of-impact sweeps — so the fast body still gets motion-clamped on a pair the user told the engine to ignore.

How

• New private helper pair_filtered_out_by_hooks in src/dynamics/ccd/ccd_solver.rs that mirrors the filter call in NarrowPhase::compute_contacts (checks ActiveHooks::FILTER_CONTACT_PAIRS on either collider, builds a PairFilterContext, returns true on None).
• Called at all three pair-evaluation sites in the CCD solver:
  • find_first_impact (substep splitting)
  • predict_impacts_at_next_positions first-sweep loop (initial TOI collection)
  • predict_impacts_at_next_positions resweep loop (after freezing bodies)
• hooks: &dyn PhysicsHooks is threaded through the two public CCDSolver methods and through PhysicsPipeline::run_ccd_motion_clamping.

Breaking change

CCDSolver::find_first_impact and CCDSolver::predict_impacts_at_next_positions now take an extra hooks: &dyn PhysicsHooks argument. PhysicsPipeline::step callers are unaffected — only code that drives the CCD solver directly (uncommon) needs to pass the hooks argument.

If you'd prefer a non-breaking additive approach (e.g., keep the current signatures and add *_with_hooks variants that take the extra param), I'm happy to restructure — I went with the breaking approach because rapier already makes breaking changes between minor versions and it keeps the surface clean.

Test

Added pipeline::physics_pipeline::test::ccd_respects_filter_contact_pair_hook (3D) that:

1. Creates a 200 m/s CCD body and a stationary CCD body directly in its path
2. Installs a hook that rejects the pair
3. Steps 5 times
4. Asserts (a) the hook was called from CCD (since narrow phase never sees the pair at these speeds), and (b) the fast body flew through to x > 10 rather than clamping near the stationary body (which happens without the fix at x ≈ -0.77).

Empirically verified on a downstream project that the same scenario fails before this patch and passes after.

CHANGELOG

Added an Unreleased section with the fix and the breaking-change note.

Notes for review

• Sensor asymmetry: the narrow phase invokes filter_contact_pair for sensor pairs too (and uses solver_flags to decide what to do downstream). CCD's sweep sites skip sensors before calling this helper, which is intentional — CCD only resolves contact TOIs, and sensor intersection events are reported elsewhere. Called out in the helper's doc comment.
• Double-filter per step: a filtered pair may get filter_contact_pair called up to 3× per step now (once per CCD sweep + narrow phase). Narrow phase already calls per-pair per-step; I don't expect the CCD sweeps to dominate in practice since CCD is only engaged for fast bodies.

[sebcrozet]

Looks good, thanks!