Add Socratic intake stage to support richer input types (fixes #16)

[yyifan-Onyen]

Summary

Closes #16

• Adds a Claude-driven Stage 00 (Research Intake) that runs before Stage 01, using the same operator + approval loop as all other stages
• Supports user-provided resources: PDFs, code repos, datasets, .bib files, notes — auto-classified and copied into workspace
• Claude analyzes the goal and resources via Socratic Q&A, producing a research brief with stage-readiness assessment
• New CLI flags: --resources PATH [PATH ...] and --skip-intake
• Fully backward compatible: --skip-intake or piped stdin preserves old behavior; all 48 tests pass (23 new + 25 existing unchanged)

What this addresses from the issue

Issue request	Implementation
Support PDFs, code, datasets, notes, bib	classify_resource() handles all types, copies to correct workspace dirs
Skip steps if resources provided	Claude's intake output includes per-stage readiness assessment
Start from actual research materials	Resources are in workspace before any stage runs
Socratic Q&A style	Claude generates context-aware questions via src/prompts/00_intake.md

Files changed

• src/intake.py — NEW: dataclasses, resource classification/ingestion, serialization
• src/prompts/00_intake.md — NEW: Claude prompt template for Socratic intake
• tests/test_intake.py — NEW: 23 test cases
• main.py — --resources, --skip-intake flags
• src/manager.py — _run_intake(), intake context injection into prompts
• src/terminal_ui.py — ask_yes_no(), ask_resource_paths(), show_intake_summary()
• src/utils.py — INTAKE_STAGE, intake_context in RunPaths

Test plan

• All 48 unit tests pass (python -m unittest discover tests/ -v)
• Integration tests: resource classification, ingestion, serialization round-trip, prompt injection
• E2E: --skip-intake backward compatibility with fake-operator
• Manual test: interactive intake with PDF resource, Claude analyzes and produces Socratic Q&A

🤖 Generated with Claude Code