[DresdenGman]

Closes #2469

Summary

As proposed in #2469, the parsing logic from the deprecated public parse_bsrn and parse_epw functions still sat in the private helpers _parse_bsrn and _parse_epw, leaving bsrn.py and epw.py structured differently from the other iotools modules and duplicating documentation. This PR moves that logic into read_bsrn and read_epw and deletes the private helpers. (The deprecated public functions were already removed in v0.16.0, so the helpers had no callers outside these two modules — 4 call sites in total.)

Changes

• pvlib/iotools/bsrn.py: moved the body of _parse_bsrn verbatim into read_bsrn; updated get_bsrn to call read_bsrn with the in-memory buffer (a path read_bsrn already supports); removed _parse_bsrn and its duplicated docstring.
• pvlib/iotools/epw.py: inlined the body of _parse_epw into read_epw; updated the HTTP branch to build a StringIO and go through _file_context_manager like the other input paths; removed _parse_epw and its duplicated docstring.
• The inlined parser bodies are verbatim apart from indentation (22 lines re-wrapped to stay within the 79-character flake8 limit) and one variable rename (fbuf → f); the dispatch around them (get_bsrn, and read_epw's HTTP branch) is what changed. Behavior is unchanged.
• docs/sphinx/source/whatsnew/v0.16.2.rst: added a maintenance entry.

Tests

• Existing: test_bsrn.py + test_epw.py gave 10 passed / 3 skipped before this change (the skips are the credential-gated remote BSRN tests) and give the same results after it.
• Behavior comparison: 7 cases (BSRN from .gz with defaults and with all longwave-radiation columns, BSRN from .dat, BSRN via a gzip buffer; EPW with defaults, with coerce_year=2015, and via StringIO) were compared before vs. after — DataFrame value hashes, shapes, and metadata fingerprints are identical in every case. Downstream checks: the Location.from_epw test passes and test_pvgis.py (which calls read_epw internally) is 29 passed.
• New mocked tests for the two rerouted entry points (4 tests, bringing the two files to 14 passed / 3 skipped): get_bsrn is exercised end-to-end against a mocked FTP connection returning the sample gzip bytes — the result is value-identical to the local read_bsrn path, and a spy check confirms read_bsrn is called exactly once with the buffer and with a non-default logical_records forwarded verbatim; read_epw is exercised through a mocked urlopen, value-identical to the local path, including forwarding of coerce_year.

[RDaxini]

Closing this since the contributing guidelines have not been followed. Will consider re-opening if compliance is achieved. Core guidelines were also highlighted to you in #2932 (see #2932 (comment)).