add torchmetrics wrapper for evaluate_boxes

[jveitchmichaelis]

Description

• Use torchmetrics to collect results and call evaluate_boxes during validation
• All metrics now respect "validate_on_epoch", but should be fast enough even with large datasets to run with n=1. Metrics are always reset regardless.
• Simpler empty frame tracking during validation_step
• Much cleaner logging in main. Non-loggable metrics are dropped in the torchmetric class.
• Use config_args to set up test fixtures instead of post-init overwriting the config. This is good practice and doing it the other way around causes undefined behaviour if certain aspects of the deepforest class aren't also updated in sync...
• ... if the user changes anything like validation data etc after initialization, we need to also re-init the metrics because precision/recall needs to know about label_dict and the annotation file when it calls evaluate under the hood.

Related Issue(s)

#901 (step towards this)
#1254
#1245
Supports #1253

AI-Assisted Development

• 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

AI tools used (if applicable):
Claude Code to do some tedious rewriting of the config_args.

[codecov]

Codecov Report

❌ Patch coverage is 97.64706% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 87.52%. Comparing base (0ab23a3) to head (8a74338).
⚠️ Report is 9 commits behind head on main.

Files with missing lines	Patch %	Lines
src/deepforest/main.py	93.10%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1256      +/-   ##
==========================================
- Coverage   87.73%   87.52%   -0.22%     
==========================================
  Files          20       21       +1     
  Lines        2716     2782      +66     
==========================================
+ Hits         2383     2435      +52     
- Misses        333      347      +14     

Flag	Coverage Δ
unittests	87.52% <97.64%> (-0.22%)	⬇️

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.

[henrykironde]

@jveitchmichaelis can you please update this PR

[jveitchmichaelis]

I might revert some of the test changes here, to keep the PR a bit more focused. We can initialize the metrics in create_trainer.