[TomCC7]

Generated pose payloads now work with Observation metadata and dimos map pose-fill. Standard nested PoseStamped, TransformStamped, PoseWithCovariance and Odometry inputs normalize to the existing indexed seven-value pose without loading or mutating the target payload. Static mounts use explicit transform composition. SQLite pose-fill preserves original CDR blobs and exact sensor Headers.

Linear: https://linear.app/dimensional/project/message-encoding-refactor-05a3ddb4af3e/overview

Parent: #4689. This layer changes pose inputs only. Observation's legacy convenience output types and their remaining consumers will be migrated together in later layers.

File changes	Files	Purpose
Runtime	2	Normalize generated pose shapes; compose mounts; scope source/destination store cleanup
Tests	2	Original matching/mount assertions plus generated identities, rotation, lazy loading, input ownership and SQLite byte preservation
CI	1	Run the same cases against installed wheel modules and strict typing

Validation: 20 focused source tests and the same 20 installed-wheel tests pass; installed production modules asserted to come from site-packages. Two production files pass scoped strict mypy; applicable hooks pass; normal wheel builds and installed dependency check passes. Wheel SHA256: 133e2dac1643019397a6d369d16d9eedaeb35fc767ee68c792442adc832d3b3e.

Manual review from this checkout after the ordinary development install:

python -m pytest --noconftest --import-mode=importlib -o addopts='' -q \
  dimos/memory/type/test_observation.py \
  dimos/mapping/cli/test_pose_fill.py \
  dimos/memory/test_stream.py::TestWithPose \
  dimos/memory/store/test_cdr_recording.py

The SQLite case writes real canonical CDR streams, applies a separately timestamped pose, reopens both databases and checks the original bytes and nanosecond Header. No historical decoding fallback, hardware, network transport or benchmark run. CI has not been polled; no whole-stack completion claim.