Implement Synix v0.1 - declarative pipeline for AI conversation exports

[marklubin]

Summary

This PR implements the complete Synix v0.1 pipeline system, including both Phase 1a (core pipeline) and Phase 1b (extended capabilities).

Phase 1a - Core Pipeline

• Two-layer storage architecture (control.db + artifacts.db)
• TransformStep (1:1 processing) and AggregateStep (N:1 with time-based grouping)
• Claude and ChatGPT export source parsers
• FTS5 full-text search with automatic triggers
• Incremental processing via materialization keys (skip already-processed records)
• CLI commands: init, run, status, search, plan, runs

Phase 1b - Extended Capabilities

• FoldStep: Sequential processing with state accumulation (like reduce/fold)
• MergeStep: Fan-in operation combining records from multiple upstream sources
• Artifact publishing: File surface for exporting results to filesystem
• CLI export command: synix export <step> [--format markdown/json/text]

Example Usage

from synix import Pipeline

pipeline = Pipeline("personal-memory", agent="mark")

# Multiple sources
pipeline.source("claude", file="exports/claude.json", format="claude-export")
pipeline.source("chatgpt", file="exports/chatgpt.json", format="chatgpt-export")

# Parallel transforms
pipeline.transform("claude-sum", from_="claude", prompt=summarize)
pipeline.transform("gpt-sum", from_="chatgpt", prompt=summarize)

# Merge multiple sources
pipeline.merge("all-summaries", sources=["claude-sum", "gpt-sum"], prompt=combine)

# Fold for evolving narrative
pipeline.fold("narrative", from_="all-summaries", prompt=evolve, initial_state="")

# Publish artifact
pipeline.artifact("report", from_="narrative", surface="file://output/report.md")

pipeline.run()

Architecture

src/synix/
├── cli.py              # Click CLI commands
├── config.py           # Pydantic settings
├── db/                 # SQLAlchemy models (control + artifacts)
├── llm/                # OpenAI-compatible LLM client
├── pipeline.py         # Core Pipeline class
├── services/           # Business logic (records, search, runs)
├── sources/            # Claude/ChatGPT parsers
├── steps/              # Transform, Aggregate, Fold, Merge
└── surfaces/           # Artifact publishing (file://)

Test Plan

• 115 tests passing (72 unit + 43 E2E)
• Linting passes (ruff check)
• Format check passes (ruff format --check)