Add MC-Dropout uncertainty estimation for patch classification

[Pier45]

Adds Monte Carlo Dropout-based uncertainty quantification (epistemic + aleatoric)
for patch classification models.

TIAToolbox's pretrained classification models (AlexNet, MobileNet, etc.) don't
expose uncertainty estimates, only point predictions. This adds a lightweight,
non-invasive wrapper to enable MC-Dropout sampling and uncertainty decomposition.

• bayesian_wrapper.py: wraps any model, injects dropout at inference if needed
• mc_dropout.py: runs N stochastic forward passes
• uncertainty.py: computes epistemic/aleatoric/total uncertainty
• Includes unit tests

[codecov]

Codecov Report

❌ Patch coverage is 93.25843% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 99.85%. Comparing base (8d13267) to head (f0e6efe).

Files with missing lines	Patch %	Lines
tiatoolbox/models/architecture/bayesian_wrapper.py	87.23%	4 Missing and 2 partials ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1104      +/-   ##
===========================================
- Coverage    99.90%   99.85%   -0.05%     
===========================================
  Files           93       96       +3     
  Lines        11911    12000      +89     
  Branches      1562     1573      +11     
===========================================
+ Hits         11900    11983      +83     
- Misses           5        9       +4     
- Partials         6        8       +2     

☔ View full report in Codecov by Harness.
📢 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.