feat(sources): add Semantic Scholar adapter with citation context

[foundatron]

Closes #3

Changes

1. tentacle/config.py

• Add SemanticScholarSourceConfig(SourceConfig) dataclass with: s2_api_key: str = "", min_citations: int = 0, days_back: int | None = None
• Change Config.semantic_scholar type from SourceConfig to SemanticScholarSourceConfig
• Add env var override in load_config: read S2_API_KEY into config.semantic_scholar.s2_api_key
• Add validation in validate: min_citations >= 0, days_back >= 1 (if set)
• Update DEFAULT_CONFIG_TEMPLATE with commented-out s2_api_key, min_citations, days_back

2. tentacle/sources/semantic_scholar.py

• Add __init__(self, api_key: str = "", min_citations: int = 0, days_back: int | None = None) — store on instance
• In _search, add x-api-key header to Request when self._api_key is non-empty
• Add citationCount to the fields query parameter
• Add publicationDateOrYear parameter (verify exact name against S2 docs) using YYYY-MM-DD:YYYY-MM-DD format when days_back is set
• Add _fetch_with_retry(self, req: Request) -> bytes method: wraps urlopen, retries on HTTP 429 respecting Retry-After header (parse as int, fallback to 1s), max 3 retries with exponential backoff. Non-retryable HTTP errors (4xx/5xx) are caught, logged, and return None
• In _search, call _fetch_with_retry instead of raw urlopen; on None return, skip that query
• After parsing, filter out papers where citationCount (defaulting None to 0) is below min_citations

3. tentacle/cli.py

• Update _get_sources to pass api_key=config.semantic_scholar.s2_api_key, min_citations=config.semantic_scholar.min_citations, days_back=config.semantic_scholar.days_back to SemanticScholarAdapter()

4. tentacle/models.py

• No changes needed

Review Findings

• Errors: 0
• Warnings: 3
• Nits: 5
• Assessment: NEEDS CHANGES

The most actionable items are: (1) catching URLError in _fetch_with_retry to handle network-level failures gracefully, (2) adding a log line when Retry-After header parsing fails, and (3) adding a test for non-integer Retry-After values. The overall implementation is solid — retry logic, config validation, env var override, and test coverage are all well-structured.