denise: score sprite/playfield collision over all enabled planes

[LinuxJedi]

Problem

Denise's sprite/playfield collision comparator (clxcon_planes_match) only
walked bitplanes up to the current BPU count. Real Denise compares every
CLXCON-enabled plane regardless of how many planes the display is fetching:
a plane enabled beyond the BPU count reads back as 0 and still gates the
match. This mirrors vAmiga's checkS2PCollisions, which evaluates
(dBuffer & enbp) == (mvbp & enbp) across all six (CLXCON) or eight
(CLXCON2, AGA) planes.

With the default CLXCON reset value (all six planes enabled, all match bits
set), a low-plane-count playfield can never satisfy the compare on hardware,
so no collision is flagged. Copperline stopped at the fetched plane count and
so reported a spurious collision, corrupting the CLXDAT value software reads.

Fix

Drop the nplanes argument from collision_pixel / clxcon_planes_match
(render path) and live_playfield_collision_pixel /
live_clxcon_planes_match (live CLXDAT path) and always iterate the full
plane bank (6 for OCS/ECS, 8 for AGA via CLXCON2).

Validation

• vAmigaTS: Denise/Sprites/collision/sprcoll1 and sprcoll1d drop from
  5.10% -> 0.12% differing pixels vs the vAmiga 4.4 reference. The other
  14 collision cases are unchanged (their residual is a different, unrelated
  class).
• Byte-identity: phooey/tomato/roots-ecs/roots-aga/zool regression
  screenshots stay byte-identical at 4s and 8s (collision affects only CLXDAT
  reads, never the rendered framebuffer).
• 1302 unit tests green, plus a new regression test
  (collision_match_gates_on_enabled_planes_beyond_the_bpu_count); clippy and
  fmt clean.

The seven existing live-CLXDAT unit tests set CLXCON to 0x0FC1 (match only
bitplane 1) so their one-plane playfields still collide under the corrected
all-planes compare, keeping their assertions intact.