[Refactor] Separate CV agent from SAM server with tool-based image analysis

[pskeshu]

Description

Refactor the SAM server to separate computer vision capabilities into a dedicated CV agent that can reason about image analysis tasks.

Current State

• sam_server.py couples SAM segmentation with visualization
• Detection pipeline: brightness → SAM → Claude verification
• No autonomous CV reasoning

Requirements

1. CV Agent Architecture

• Receives image + intent from orchestrator
• Autonomously selects appropriate CV tools
• Returns structured results with confidence

2. Tool Categories

• Classical CV: thresholding, morphology, edge detection, blob analysis
• SAM Tools: segmentation, mask refinement, point/box prompts
• VLM Tools: Claude Vision queries, object detection prompts
• Measurement: area, intensity, shape metrics

3. Agent Capabilities

• Multi-step analysis (e.g., "find dim embryos" → enhance contrast → detect → filter by brightness)
• Explain reasoning in results
• Suggest alternative approaches if initial method fails

4. Clean SAM Server

• Pure segmentation service (stateless)
• Remove visualization code
• Simple API: image + prompt → masks

Technical Approach

• Create gently/agent/cv_agent.py with tool-using Claude instance
• Move classical CV to gently/cv/ module
• Simplify backend/sam_server.py to pure SAM inference
• CV agent uses SAM server as one of its tools

Key Files

• backend/sam_server.py
• gently/agent/sam_detection.py
• New: gently/agent/cv_agent.py
• New: gently/cv/classical.py, gently/cv/measurement.py