Fix camera type dispatch in 3D demos

[ziyangwang007]

Motivation

The --cam-type argument in mono_det_demo.py and
multi_modality_demo.py is currently removed from call_args before calling
the inferencer. As a result, the user-provided camera type is not propagated to
the preprocessing stage.

In addition, Base3DInferencer._dispatch_kwargs consumes cam_type but only
uses it to set cam_type_dir. It does not dispatch cam_type back to
preprocess_kwargs, so Base3DInferencer.__call__ falls back to the default
CAM2.

This can cause demo failures such as KeyError: 'CAM2' even when users pass
a valid camera type like CAM0, CAM_BACK, or CAM_FRONT.

Fixes #2881.
Fixes #2868.
Related to #2981.

Modification

• Keep cam_type in call_args in demo/mono_det_demo.py.
• Keep cam_type in call_args in demo/multi_modality_demo.py.
• Dispatch cam_type to preprocess kwargs in
  Base3DInferencer._dispatch_kwargs.
• Continue using the same cam_type value for cam_type_dir, so saved
  visualization directories remain consistent with the selected camera.

BC-breaking (Optional)

No backward-incompatible change is expected. The change makes the existing
--cam-type argument take effect as intended.

Use cases (Optional)

This fixes demo commands that explicitly specify a camera type, for example:

python demo/multi_modality_demo.py \
  demo/data/sunrgbd/000017.bin \
  demo/data/sunrgbd/000017.jpg \
  demo/data/sunrgbd/sunrgbd_000017_infos.pkl \
  configs/imvotenet/imvotenet_stage2_8xb16_sunrgbd-3d.py \
  checkpoints/imvotenet_stage2_16x8_sunrgbd-3d-10class_20210819_192851-1bcd1b97.pth \
  --cam-type CAM0

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}