Fix SITL vertical anchor altitude bug; add EK3 baro/GPS altitude blending

[mcroomp]

Summary

Fixes a SITL crash in test_lua_flight_steady_sitl:
rawes.lua:315: bad argument #2 to 'set_target_angle_and_rate_and_throttle' (out of range).

Root cause: ArduPilot's vertical position estimate is baro-anchored — whatever
altitude the vehicle physically occupies when the EKF's local origin is
established becomes its "zero" (real baro-sensor behavior, not a simulation
artifact). This SITL test boots the vehicle already kinematically locked
ctx.home_alt_m metres above the anchor, so the anchor's sent altitude
(RAWES_AAL, via NAMED_VALUE_INT) needs to be offset by -ctx.home_alt_m to
resolve to the correct (positive-down) D offset once EKF-local. Without this
offset, the elevation-angle computation (asin(-rz/tlen)) was drastically
wrong, driving the cyclic pitch command to saturate until it exceeded
ArduPilot's valid range.

Diagnosed by comparing telemetry.csv's ground-truth pos_z (constant, correct)
against ekf_pos_z (stuck near 0 throughout) — confirming ArduPilot's own
vertical EKF estimate never reflected the true altitude.

Changes

• simulation/tests/sitl/flight/conftest.py: fix _send_anchor_location()
  to send anchor_alt_m = HOME_ALT_M - ctx.home_alt_m instead of the raw home
  altitude. Also includes general cleanup (unused numpy import removed, a
  stale log message corrected, minor blank-line/indentation fixes).
• simulation/scripts/rawes.lua: adds throttled (~1 Hz) pre-capture
  diagnostic logging of pos_ned/anchor geometry, and expands the "captured"
  STATUSTEXT with explicit geometry values — a permanent diagnostic tool for
  future geometry-capture bugs.
• simulation/tests/sitl/rawes_common_defaults.parm: sets
  EK3_OGN_HGT_MASK=5 (bits 0+2: correct when Baro is the active height
  source, apply the correction to local position) so GPS altitude
  progressively corrects baro-drift bias in the EKF's local vertical position
  over long real-world flights — mitigates weather-driven QNH drift, which
  the one-time-resolved anchor altitude can't otherwise self-correct.
• Remaining files (gcs.py, mediator.py, mock_ardupilot.lua,
  rawes_lua_harness.py, rawes_modes.py, stack_infra.py, stack_utils.py,
  analysis/*.py, various simtests/unit tests, design docs, AGENTS.md)
  carry the broader anchor-location-over-GPS migration this fix depends on:
  transmitting the tether anchor's absolute GPS location
  (RAWES_LAT/RAWES_LON/RAWES_AAL via NAMED_VALUE_INT) instead of
  assuming a fixed NED offset, so the design works with arbitrary real-world
  GPS coordinates rather than only a fixed simulated origin.

Validation

bash test.sh stack -n 1 -k test_lua_flight_steady_sitl → PASSED
(stable=128s, max_activity=423 PWM), confirmed both before and after adding
EK3_OGN_HGT_MASK=5.

Pumping and landing SITL stack tests have not yet been re-validated with these
changes (next step per AGENTS.md).