Add 2D input support for detection with clear guardrails

[Aayannoori15]

Before submitting a pull request (PR), please read the contributing guide.

Please fill out as much of this template as you can, but if you have any problems or questions, just leave a comment and we will help out :)

Description

What is this PR

• Bug fix
• [yes] Addition of a new feature
• Other

Why is this PR needed?
To allow 2D microscopy inputs to run through the detection pipeline safely. This unblocks users working with single‑plane data and provides clear guardrails around unsupported 2D classification.

What does this PR do?
Accepts 2D inputs for detection and treats them as a single z‑plane.
Forces z‑dependent filter parameters to 1 in 2D mode.
Pads voxel_sizes from (y, x) to (1.0, y, x) for 2D.
Adds guardrails and clearer errors in the main pipeline and classifier entry point for 2D usage.

Please reference any existing issues/PRs that relate to this PR.

How has this PR been tested?

None yet

Please explain how any new code has been tested, and how you have ensured that no existing functionality has changed.

Is this a breaking change?

No. Existing 3D workflows are unchanged. 2D inputs now have explicit validation.

If this PR breaks any existing functionality, please explain how and why.
No. Existing 3D workflows are unchanged. The new logic only activates for 2D inputs and adds explicit validation/guardrails.

Does this PR require an update to the documentation?

Yes. Usage docs should mention that 2D detection is supported but classification is not yet implemented for 2D.

If any features have changed, or have been added. Please explain how the documentation has been updated (and link to the associated PR). See here for details.

Checklist:

• [yes ] The code has been tested locally
• [no ] Tests have been added to cover all new functionality (unit & integration)
• [ no] The documentation has been updated to reflect any changes
• [ yes] The code has been formatted with pre-commit