[TomCC7]

Generated TF timestamps cannot pass through f64 seconds without losing epoch nanoseconds. The native SDK now retains optional exact stamp metadata through buffering, latest/reverse/composed lookups and local publication. Samples colliding as floats are ordered by their retained source nanoseconds. Existing Transform::new, public ts, lookup signatures and seconds-based math remain available; changing ts invalidates retained metadata.

This is the prerequisite for the native navigation CDR adapters. It does not change default encoding factories or the float-based lookup query/tolerance API.

Parent: #4680. Project: https://linear.app/dimensional/project/message-encoding-refactor-05a3ddb4af3e/overview

Category	Files	Change
SDK + adjacent tests	1	Exact metadata and five behavior tests; existing 37 TF tests retained
CI	1	Focused TF tests and all-target Clippy

From this checkout, run:

cargo test --locked --manifest-path native/rust/dimos-module/Cargo.toml \
  --features cdr --lib --jobs 2 tf::tests

Using the installed main wheel's existing SDK Cargo export, the new API is:

use dimos_module::{Transform, nalgebra::Isometry3};
let stamp = 1_700_000_000_123_456_789;
let edge = Transform::from_nanoseconds("world", "base", stamp, Isometry3::identity());
assert_eq!(edge.stamp_nanoseconds(), Some(stamp));
assert_eq!(Transform::new("world", "base", 1.25, Isometry3::identity()).ts, 1.25);

Validation: 42 TF tests pass, all-target Clippy with warnings denied passes, other applicable hooks pass, normal main wheel builds and the full runtime environment passes pip check (198 packages). Installed exported tf.rs SHA256 matches the checkout; an actual separate Cargo project outside this checkout compiled and ran against that installed SDK, reporting exact source nanoseconds 1700000000123456789. No repository-relative source dependency was used by that consumer.

Nix lint was attempted and failed because nix is absent; it is not counted as passing and the repository gate remains intact. No hardware, benchmark or GitHub CI result is claimed. Native navigation codec adoption and its integration checks are the next layer.