Add Metric Depth Estimation (DAv3)

[mrpositron]

What has changed and why?

Adds metric depth estimation for Depth Anything V3. The new DepthAnythingV3MetricDepthEstimation task model (dinov2/dav3-metric-large) reuses the relative model's backbone and DPT decoder, and scales the output to metric units via metric_depth = focal * output / 300 when camera intrinsics are given. dpt.py is moved to the shared depth_estimation_components/ package, and convert_checkpoint.py now also converts the official DA3METRIC-LARGE weights.

How has it been tested?

NYUv2 Benchmark:

Model	a1 ↑	a2 ↑	a3 ↑	abs_rel ↓	log_10 ↓	rmse ↓	rmse_log ↓	silog ↓	sq_rel ↓
Official Implementation	0.948	0.984	0.992	0.08	0.036	0.342	0.116	9.112	0.047
LightlyTrain Port	0.948	0.984	0.992	0.08	0.036	0.342	0.116	9.103	0.046

Did you update CHANGELOG.md?

• Yes
• Not needed (internal change)

Did you update the documentation?

• Yes
• Not needed (internal change without effects for user)

[mrpositron]

@codex

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Nice work!

Reviewed commit: 0fdc355e6a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[mrpositron]

/review