Add Claude and Gemini shared chat extraction

[adventuremommy]

Summary

• add structured shared-chat extraction for Claude public links, including Next.js __NEXT_DATA__ payloads and existing preloaded state payloads
• expand Gemini public-share DOM selectors and add structured-script fallback
• report known private/missing provider pages as unavailable instead of falling through to generic extraction behavior
• declare python-multipart, which FastAPI needs to import routes that use File(...)

Fixes #155.

Verification

• uv run --extra dev pytest tests/test_api_memory_scrape.py -q passes
• uv run --extra dev ruff check src/api/routes/memory.py tests/test_api_memory_scrape.py pyproject.toml passes

Repository-wide checks currently have unrelated baseline failures:

• uv run --extra dev pytest fails in tests/test_enterprise_chat.py::test_annotation_service_extracts_and_stores_project_annotations because EnterpriseAnnotationService.extract_and_store() returns (['ann_1'], '') while the test expects ['ann_1']; this also fails when run by itself.
• uv run --extra dev ruff check . reports existing lint errors in files outside this PR, including server.py, src/api/dependencies.py, and scanner modules.
• uv run --extra dev mypy src || true stops on an existing duplicate module name between src/prompts/judge.py and src/prompts/examples/judge.py.

[gemini-code-assist]

Code Review

This pull request refactors the chat extraction logic for ChatGPT, Claude, and Gemini by introducing modular helper functions for DOM and script-based parsing, and adds a check for unavailable conversations. It also includes a new test suite and adds the python-multipart dependency. Feedback was provided to address a fragile regex that could truncate JSON data and to optimize HTML normalization for better performance.

[gemini-code-assist]

The non-greedy regex (\{.*?\}); is fragile because it will terminate at the first occurrence of };. If the chat content contains code snippets or strings that include };, the extraction will result in truncated, invalid JSON. Using a greedy match to find the last closing brace is more robust for these state-carrying scripts.

Suggested change

	match = re.search(
	r"__PRELOADED_STATE__\s*=\s*(\{.*?\});",
	script_text,
	re.DOTALL,
	)
	match = re.search(
	r"__PRELOADED_STATE__\s*=\s*(\{.*\})",
	script_text,
	re.DOTALL,
	)

[gemini-code-assist]

Normalizing the entire HTML string using " ".join(html.lower().split()) is very inefficient for large documents, as it creates multiple large intermediate objects (a lowercased string, a list of all words, and a new joined string). Since the markers are simple phrases, searching directly in the lowercased HTML string is significantly more performant and sufficient for this check.

Suggested change

	lowered = " ".join(html.lower().split())
	lowered = html.lower()

[ishaanxgupta]

Hi @adventuremommy thank you for the contribution, please have a look on the gemini suggestions