Description

Fixes #3857.

cape_cin applies the virtual temperature correction by replacing the environmental
and parcel temperature profiles with their virtual temperatures before calling lfc
and el:

temperature = virtual_temperature_from_dewpoint(pressure, temperature, dewpoint)
parcel_profile = virtual_temperature(parcel_profile, parcel_mixing_ratio)
lfc_pressure, _ = lfc(pressure, temperature, dewpoint, parcel_temperature_profile=parcel_profile, ...)
el_pressure, _ = el(pressure, temperature, dewpoint, parcel_temperature_profile=parcel_profile, ...)

lfc and el then recompute the LCL from those arrays — lfc from
parcel_temperature_profile[0], el from temperature[0] — i.e. they treat the
virtual temperature as if it were the actual temperature. Because virtual temperature
is always ≥ temperature, the LCL comes out at too low a pressure (too high an
altitude). Since the LFC and EL are floored at the LCL, both get biased upward; the
effect is strongest on the LFC, exactly as described in the issue.

Fix

Give lfc and el optional lcl_pressure (and, for lfc, lcl_temperature)
keywords so the caller can supply an LCL computed from the actual temperatures. When
the keywords are omitted the LCL is computed from the inputs exactly as before, so
direct callers and the public API are unchanged.

cape_cin already computes the LCL from the actual temperatures (for its below_lcl
mixing-ratio logic) before the virtual-temperature substitution, so it simply passes
that LCL through to lfc/el.

Effect

For a sounding whose LFC is floored at the LCL, the actual-temperature LCL sits about
30–45 hPa higher in pressure than the virtual one, so the LFC (and the CAPE integration
that starts there) is corrected accordingly. For example, on the profile in
test_cape_cin_value_error the surface-based CAPE goes from the previously-reported
2161.9 J/kg to 2206.7 J/kg once the LFC is floored at the correct LCL.

Tests

• New test_lfc_el_lcl_from_actual_temperature: reconstructs the virtual-temperature
  profiles cape_cin builds and shows that, without the override, lfc floors at the
  spuriously-low virtual LCL, while supplying the actual-temperature LCL floors it at
  the correct LCL (and that el honors the keyword).
• Updated the expected CAPE in test_cape_cin_value_error (whose LFC is floored at the
  LCL) with a comment explaining the change.

The full tests/calc/test_thermo.py suite passes (228 tests); ruff is clean.