docs(kinematic): document units for all WRAP accessors

[TommyC81]

What & why

The WRAP kinematic accessors in openap/kinematic.py return values in a mix of km, m/s, and m/s², but only climb_range() documented its unit. Every other accessor had a unit-less docstring (e.g. "Get takeoff distance."), so a consumer assuming meters / knots / ft·min⁻¹ silently gets a value off by 10–1000×.

The WRAP data is internally consistent — this is a documentation fix, not a computation change. Closes #69.

from openap import WRAP
w = WRAP("a320")
w.takeoff_distance()["default"]  # 1.65  -> was "1.65 of what?" — actually km
w.cruise_alt()["default"]        # 10.92 -> actually km (≈ FL358)
w.takeoff_speed()                # 85.3  -> actually m/s (166 kt)

Changes (docstrings only — zero logic change)

• Added the unit to 27 accessors (km for distances/altitudes, m/s for speeds & vertical rates, m/s² for accelerations), matching the existing climb_range() (in km) style.
• Added the 3 missing docstrings (cruise_max_alt, cruise_max_mach, cruise_mean_vcas).
• Fixed landing_distance() typo: "breaking" → "braking".
• Disambiguated a duplicate docstring: descent_cross_alt_concas() was a verbatim copy of …conmach(); now reads "from constant CAS to post-constant-CAS descent", matching the descent_vs_post_concas() naming already in the file.

Two minor rewordings beyond a pure unit-add, flagged for review:

• climb_const_mach / descent_const_mach: "speed during constant Mach …" → "Mach number during constant Mach …" (Mach is dimensionless; there is no unit to append).
• descent_cross_alt_concas: "from constant CAS to post-constant-CAS descent" — the constant-CAS phase ends at the lower crossover (de_h_cas_const ≈ 5.7 km), below which the descent decelerates toward approach. Verified against the three descent vertical-rate variables (de_vs_avg_mach_const / …cas_const / …after_cas) which encode the three-phase structure, and against the descent_vs_post_concas() method name.

Verification

No return values, signatures, or imports are touched. Confirmed by stripping all docstrings and diffing the remaining code — original and working copy are identical except for blank lines where docstrings were added to the three previously-bare functions.