BUG: Convert grad/radian angular parameters to degrees in CRS.to_cf

[charan-rathore]

Fixes #1641.

Problem

CRS.to_cf() wrote angular parameter values (prime meridian longitude,
longitude/latitude of natural origin, standard parallel) in whatever unit
the CRS defines them in. For CRSs with grad angular units such as the
French Lambert zones (EPSG:27571/27572/27573, NTF Paris datum), the CF dict
contained values like standard_parallel = 52.0 and
longitude_of_prime_meridian = 2.5969213 - grad numbers in fields that CF
expects to be degrees. Round-tripping through CRS.from_cf then
reinterpreted those numbers as degrees and the reprojection was off by
hundreds of kilometers.

Fix

Added a shared _to_degrees() helper in pyproj/crs/_cf1x8.py that
converts grad and radian values to degrees using the parameter's unit
conversion factor, leaving degree values untouched. CRS._to_dict() now
routes angular parameters through it, and CRS.to_cf() uses it for
longitude_of_prime_meridian.

Tests

Three regression tests in test/crs/test_crs_cf.py:

• test_to_cf__grad_units__converts_to_degrees: EPSG:27572 exports
  standard_parallel 46.8 and longitude_of_prime_meridian 2.33722917.
• test_to_cf__grad_prime_meridian__converts_to_degrees: +proj=longlat +pm=paris exports the Paris prime meridian in degrees.
• test_cf_round_trip__grad_units: the from_cf round trip keeps the
  natural-origin latitude in degrees and lands within 10 m of the direct
  transform (previously ~579 km off).

Also added the change to docs/history.rst under Latest.

Verification

run_tests.sh clones the repo at the base, confirms the patch applies,
installs the released pyproj 3.8.0 wheel (its crs.py and _cf1x8.py are
byte-identical to the base commit, so the wheel exercises the same code the
patch changes), and runs master's test/crs/test_crs_cf.py against the
installed wheel:

• Baseline (unpatched): the 3 grad-unit tests fail.
• Patched (patch applied to the installed files): all 3 pass, and the rest
  of the suite shows zero new failures compared with baseline.

Caveats

• The suite has 19 pre-existing failures in this setup, identical before
  and after the patch. They come from running master's tests against the
  3.8.0 wheel's bundled PROJ rather than a source build, not from this
  change.
• Residual round-trip error is ~3 m, not exact: to_cf does not write any
  scale-factor entry for this 1SP Lambert projection, so from_cf defaults
  scale_factor_at_natural_origin to 1.0 instead of the original
  0.99987742. That is a separate pre-existing gap (a missing CF field, not
  a units bug) and is left out of scope here; the round-trip test asserts
  within 10 m and documents this.
• The round-trip improvement was verified on the released wheel with
  bundled PROJ 9.8.1; a source build against the system's PROJ was not run.