feat(exa): add attribution header and expose full search surface

[tgonzalezc5]

Summary

Enhances the existing Exa search backend (src/fim_one/core/web/search/exa.py) to close gaps against the current Exa Search API reference:

• Sends x-exa-integration: fim-one on every request so API usage is attributed back to the FIM One integration in Exa's maintainer view (attribution only; no per-user data).
• Supports all current search types (auto/neural/fast/deep-lite/deep/deep-reasoning/instant) via EXA_SEARCH_TYPE.
• Requests highlights + text together (and optional summary), and cascades the result snippet through whichever fields the API populates, matching the Exa docs' multi-content semantics.
• Exposes Exa's filtering surface: category, includeDomains, excludeDomains, startPublishedDate, endPublishedDate, and maxAgeHours — all configurable via constructor args or env vars so get_web_searcher() still selects Exa generically.
• Validates search_type and category on construction (fail fast on typos).
• Documents the new knobs in example.env.

Fully backwards-compatible: default behaviour (type: "auto", text + highlights, 800-char snippets) is a strict upgrade over the previous text-only mode, and ExaSearch(api_key=...) keeps working unchanged.

Example

from fim_one.core.web.search import ExaSearch

searcher = ExaSearch(
    search_type="neural",
    category="research paper",
    include_domains=["arxiv.org", "nature.com"],
    max_age_hours=24 * 30,
)
results = await searcher.search("superconductor breakthrough", num_results=10)
# results[i].snippet cascades: highlights -> text -> summary

Or entirely via env:

EXA_API_KEY=exa_...
EXA_SEARCH_TYPE=fast
EXA_CATEGORY=news
EXA_INCLUDE_DOMAINS=nytimes.com,reuters.com
EXA_MAX_AGE_HOURS=24

Files changed

• src/fim_one/core/web/search/exa.py — payload builder, snippet cascade, env-driven config, attribution header
• tests/test_web_providers.py — unit tests for snippet cascade, payload shape, env var resolution, attribution header on the wire
• example.env — document new EXA_* variables

Test plan

• uv run ruff check src/fim_one/core/web/search/exa.py tests/test_web_providers.py — passes
• uv run mypy src/fim_one/core/web/search/exa.py — passes
• uv run pytest tests/test_web_providers.py — 50 tests pass (18 new, all green)
• uv run pytest tests/ --ignore=tests/test_sandbox_backend.py — 3060 tests pass, no regressions (Docker sandbox suite requires a local Docker daemon and was not run)
• Manual smoke test against live Exa API with a real EXA_API_KEY (not run in this environment; recommend a maintainer verify)

[tao-hpu]

Thanks for the contribution, @tgonzalezc5! Merged.

The integration attribution header and the full search-type surface (including deep-reasoning and instant) are both nice upgrades for our users — appreciate you closing the gap against the latest Exa Search API reference in one pass.

[tao-hpu]

@tgonzalezc5 — noticed DMs are closed on X. If there's an email or Slack channel for partnership-related follow-ups (OSS/startup credits, Featured Integration listing, co-marketing, etc.), happy to take it there. Otherwise no worries — the integration ships either way.