PTZApp: Florence-2 detection enhancements and plant identification extension

[saumya-pailwan]

This pull request introduces the extension of PTZApp, developed as part of the Summer 2025 project.
It significantly enhances the existing PTZ-YOLO system, transforming it from a static detection pipeline into a context-aware, self-directing observation system.

Key Highlights

• New Intelligent Cascade Workflow
  Adds scene captioning and contextual grounding using Florence-2 for dynamic, informed scanning.

• Enhanced Model Support
  Supports both YOLO (fast predefined detection) and Florence-2 (vision-language scene understanding).

• Example Application: PlantNet Integration
  Demonstrates an optional use case for species identification using the sharpest image from best-of-N captures.
  (Note: PlantNet is provided as an example application and is not required for PTZ-YOLO core functionality.)

• Best-of-N Capture & Blur Gate
  Automatically selects the clearest image for analysis and retries if all are blurry.

• Comprehensive Metadata Logging
  Publishes detailed telemetry: scene captions, detections, PlantNet results, and alert streams.

• Alert System
  Detects target or invasive species via configurable JSON lists and emits high-priority alerts.

• Edge-Ready Deployment
  Fully containerized for Jetson and GPU servers, supporting offline inference.