Add read_gifts as one entry point over the gift bridge presets

[shivamlalakiya]

Summary

CiviCRM, Raiser's Edge and NPSP each ship their own read_<source>_... loader and <source>_..._to_features aggregator pair. That's fine for a caller who only ever works with one CRM's export, but a caller working with more than one CRM export format (a consultant with clients on different systems, a tool that takes "whichever export you have") ends up importing three differently named function pairs to do the same read-then-aggregate operation.

read_gifts(path_or_df, *, source=...) looks that pair up in a small registry (GIFT_SOURCES = ("civicrm", "raisers_edge", "npsp")) instead, and runs it:

read_gifts("gifts.csv", source="npsp")
# equivalent to:
npsp_opportunities_to_features(read_npsp_opportunities("gifts.csv"))

It also accepts an already-in-memory DataFrame or iterable of mappings (skips the read step), and passes any extra keyword arguments straight through to the matched aggregator, so exclude_stages, exclude_gift_types and statuses all still work per-source.

Design choice: kept civicrm_contributions_to_features where it already lives in _civicrm.py, rather than physically relocating it into a new neutral module. Raiser's Edge and NPSP already import it cleanly as their shared aggregator; the "extraction" this PR delivers is uniform access through one function name rather than three, which is the lower-risk, smaller change and needed no edits to _civicrm.py, _raisers_edge.py, or _npsp.py. cli.py's existing three-way --source dispatch in _cmd_features is left untouched: it does its own per-step error handling (a separate try around the read and the aggregate step, each raising a different CLI-friendly message) that routing through read_gifts would have to either duplicate or change, and this PR's job is the library function, not a CLI refactor.

Test plan

• python -c "from philanthropy.ingest import read_gifts, GIFT_SOURCES; print('OK')" before writing tests
• New tests/test_read_gifts.py: one test per preset showing read_gifts(path, source=X) matches the two-step reader+aggregator call, a DataFrame-input test, an iterable-of-mappings test, an unknown-source ValueError test, and a kwarg-passthrough test (exclude_stages=None reaching the NPSP aggregator)
• tests/test_civicrm.py, tests/test_raisers_edge.py, tests/test_npsp.py, tests/test_cli.py all pass unmodified
• make ci: 2148 passed, 8 skipped, 98.58% coverage (floor 92%)
• make riskcov: 98% coverage over the risk tier (floor 93%)
• CHANGELOG.md entry under [Unreleased] -> Added
• docs/reference/index.md Tier 2 (Beta) row for read_gifts / GIFT_SOURCES