Feat/validator transcriber docker fix

[Acuspeedster]

Closes #114
Closes #115
Closes #116

Summary

This PR introduces two major feature enhancements and resolves a Docker environment bug:

• 🚀 Schema validation + error recovery for LLM output
• 🚀 Voice transcription pipeline via faster whisper
• 🐛 Docker PYTHONPATH consistency fix

All changes are fully tested and backward compatible.

---

🚀 Feature 1: Schema Validation

Adds SchemaValidator to validate and classify extracted LLM output:

• Detects missing or invalid fields
• Attempts safe type coercion
• Produces structured ValidationReport
• Prevents silent PDF corruption
• Surfaces operator warnings

Improves robustness of the AI → JSON → PDF pipeline.

---

🚀 Feature 2: Voice Transcription

Implements fully local transcription using faster-whisper:

• New Transcriber class
• POST /transcribe/ endpoint
• Supports common audio formats
• Lazy import handling
• No external API calls
• Fully Docker compatible

Completes voice → JSON → PDF workflow.

---

🐛 Fix: Docker PYTHONPATH

Standardizes:
PYTHONPATH=/app

instead of /app/src.

Ensures consistent import behavior across:

• Local development
• Docker runtime
• Test environments

---

Test Coverage

• Validator tests added
• Transcriber tests added
• All existing tests passing
• Verified locally and in Docker

---

Impact

• Improved extraction reliability
• Completed advertised voice feature
• Stable Docker environment
• Increased production readiness
• Fully aligned with FireForm sustainability goals