[AI-AGENT] Replace hardcoded draft amounts with structured NL→intent parsing

[wheval]

Feature Description

draft-intent returns hardcoded amount '10' (#932). Implement real parsing (rules or LLM behind flag) producing typed intents with risk scores (#837 done).

Problem Statement

AI agent MVP cannot draft correct payments; misleading for integrators.

Proposed Solution

1. Deterministic parser for amount/asset/destination patterns.
2. Optional LLM provider behind env key (never auto-execute).
3. Property tests for amount extraction.
4. Close [AI-AGENT] draft-intent returns hardcoded amount '10' regardless of user input #932.

Impact Area

• Infrastructure
• SDK

Priority

• Medium

Implementation Complexity

• Complex

Related

#932, roadmap 5.9

[Okorie2000-code]

Hi Maintainer,

I'd like to work on this issue. I have experience with TypeScript, backend development, and API design. I can submit a clean PR within 48 hours.

[funds0033-cmyk]

I will like to work on this

[nurudeenheroic]

Hello! I'd appreciate the opportunity to work on this issue.