Improve CLI and separate into sub-scripts

[jveitchmichaelis]

Description

This is a cleaned up version of the training script I used for LIDAR pre-training. I took some inspiration from the bird branch as well to add some more comet features. Docs are updated for all three scripts.

My goal is that users should never need to write scripts and we can be opinionated about defaults that we commonly use ourselves.

• Split tools into sub scripts for easier organization and testing
• Improvements to the basic training script to support common training situations:
  • Comet/Tensorboard logging, optional
  • Sensible defaults for log directories, including using comet's experiment key if generated/set and tags
  • Defaults for callbacks (image + checkpoints)
  • Resume support
  • Optional tracing
• Predict script that supports single/tiled/CSV
• Evaluate script that takes a prediction CSV from above, or runs on validation.csv_file
• Better test coverage
• Reset the default log root directory to ./lightning_logs (as opposed to $PWD).
• Small fixup in checkpoint tests where we were using a non-standard key for the log directory

AI-Assisted Development

Some use of Claude Code for assisted testing (e.g. checking it can start training jobs).

• I used AI tools (e.g., GitHub Copilot, ChatGPT, etc.) in developing this PR
• I understand all the code I'm submitting
• I have reviewed and validated all AI-generated code

[codecov]

Codecov Report

❌ Patch coverage is 67.19577% with 62 lines in your changes missing coverage. Please review.
✅ Project coverage is 86.95%. Comparing base (94af652) to head (1d7e2a9).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
src/deepforest/scripts/cli.py	0.00%	29 Missing ⚠️
src/deepforest/scripts/train.py	61.64%	28 Missing ⚠️
src/deepforest/scripts/predict.py	90.32%	3 Missing ⚠️
src/deepforest/scripts/evaluate.py	96.22%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1255      +/-   ##
==========================================
- Coverage   87.16%   86.95%   -0.21%     
==========================================
  Files          21       24       +3     
  Lines        2851     3005     +154     
==========================================
+ Hits         2485     2613     +128     
- Misses        366      392      +26     

Flag	Coverage Δ
unittests	86.95% <67.19%> (-0.21%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jveitchmichaelis]

This is squashed and ready for review. These are new public features so feel encouraged to try them out and we can tweak behavior to suit.

[bw4sz]

I agree. There are some options that I never personally use, like tensorboard, but I don't think that should limit merging. The whole point here is that the user is not looking at the script, they are using the CLI, so I care less about the large amount of code, since the user doesn't engage with it directy.