feat: custom TensorFlow.js model for waste contamination detection (#673)

[knoxiboy]

Resolves #673

Description

This PR integrates a custom TensorFlow.js classification model inside the AI Bio Scanner to detect bio-waste contamination (plastics, metal, glass) and output a granular organic purity score, automatically flagging contaminated batches.

What changed

• Added loadCustomModel and runCustomModel inside src/scanner.js using tf.sequential to define a lightweight CNN layer structure.
• Modified classifyResult to accept and process probabilities from the custom TF.js model.
• Established a threshold (90% organic purity) below which batches are automatically flagged as contaminated.
• Updated src/vision-scanner.js to utilize the custom TF.js model instead of simulated/mock values, maintaining alignment.
• Exposed loadCustomModel in BioScanner public API.

Why

Standard MobileNet can classify object labels but cannot quantify contamination percentages or detect if non-biodegradable items are mixed within the organic material, which is critical for anaerobic digestion safety.

How to test

1. Run npm run test to verify structure health.
2. Load index.html in browser.
3. Open the AI Bio Scanner, upload a waste image, and verify that the organic content meter displays the granular percentages and flags batches containing high plastic/metal/glass contamination.

Checklist

• Tested changes locally
• No lint/syntax errors introduced
• Matches all issue criteria

Files Affected

• src/scanner.js
• src/vision-scanner.js

[knoxiboy]

@Shruti070107 Please review this pr and merge it