Closes the v10.11.0 known limitation. The batched bridge SPI now
handles K>0 partial-resolution and ROI cases inside
JP3DSliceStackCodec via a new JP3DBridgeOptions overload — the
v10.11.0 architecture extends naturally with a single options
struct that rides through the bundle.

JP3D full-decode + partial-res + ROI wins (M2 release, --jp3d
bench, in-process, 7 runs / 2 warmups, median):

  Fixture          vx        full Δ ms    res1 Δ ms    ROIq Δ ms
  --------------------------------------------------------------
  mr_3d_small      262K      −0.05        +1.86        −0.02
  ct_3d_small      1.05M     **−4.19**    −0.32        −2.19
  us_3d_small      1.84M     **−4.69**    +1.25        −2.73
  mr_3d_mid        2.10M     **−9.35**    −1.63        **−4.22**
  ct_3d_mid        8.39M     **−38.85**   **−10.44**   **−21.18**
  ct_3d_large      16.78M    **−82.89**   **−25.82**   **−47.53**

Per acceptance discipline (≥3 ms wins): full 4/6, res1 2/6,
ROIq 3/6 fixtures clear. Smaller fixtures wash on the lighter
lanes (reduced workload leaves less to amortise). Both 8M+ CT
volumes (radiologist scrolling case) win on every lane.

Architectural surface — new opaque options:
  • JP3DBridgeOptions(partialResolutionLevel:, regionOfInterest:)
  • J2KDecoder._jp3dDecodeToCoefficients(_:Data, options:JP3DBridgeOptions)

Pipeline:
  • DecoderPipeline.decodeToCoefficients(_:Data, options:) sets
    partialResolutionLevel + regionOfInterest before extractTileData,
    captures options into the returned bundle.
  • iDWTAndFinalizeCoefficients honours captured options:
    partial-res → sets outputDimensions for reconstructImage;
    ROI → post-iDWT crop via Self.cropImage.
  • iDWTAndFinalizeCoefficientsBatched eligibility gate now accepts
    uniform partial-res K + ROI across the batch. Orchestrator
    truncates the chain to effectiveLevels = N − K for partial-res;
    runs full chain for ROI then per-slice crops. K=0 thumbnail
    falls back to per-slice serial (no iDWT chain to dispatch).

JP3DSliceStackCodec wiring:
  • Removes the (K>0 || ROI) → per-slice-serial fallback in the
    bulk batched path. Both branches now build a JP3DBridgeOptions,
    decode all slices to coefficients in parallel via the new
    overload, then run ONE batched iDWT + finalize.
  • The (K>0 AND ROI) compositional throw still fires (2D codec
    doesn't compose those — separate arc).
  • Env-var off-switch J2K_JP3D_BATCHED_BRIDGE=0 still works.

Validation:
  - V10_21_BatchedBridgeOptionsParityTests   7/7  PASS
    (K=0/2/4/N, ROI 128² and 64², K+ROI composition)
  - V10_20_BatchedBridgeParityTests          5/5  PASS
  - V10_20_BatchedInverseInt32ParityTests   12/12 PASS
  - V10_20_JP3DBridgeParityTests             5/5  PASS
  - `swift test --filter JP3D` regression  519/519 PASS
  - Mandatory commit gate (release mode)     7/7  PASS

Codestream bytes byte-identical to v10.11.0; encoder unchanged.
MINOR per RELEASING.md — additive public surface, no removals,
no default flip that affects bytes.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>