Fix misleading extrapolation time in buffer_core (#832) (backport #896)

[mergify]

Description

This PR fixes the issue where extrapolation errors incorrectly prioritize metadata gathered during the initial source-to-root traversal (Phase 1). By comparing timestamps from both traversal phases, the error now correctly reports the 'latest data' time for the specific link on the requested transform path.

Verification

Reproduced with the script in #832:

• Before: Reported 10s staleness from unrelated 'map' frame.
• After: Reports 0.1s staleness for the actual 'odom -> base_link' path.

Fixes #832

---

This is an automatic backport of pull request #896 done by Mergify.

[ahcorde]

Pulls: #899
Gist: https://gist.githubusercontent.com/ahcorde/31bed31c844555d7e425aee6faff29d6/raw/0b4a42679888c861c7f52263262809634ea6913e/ros2.repos
BUILD args: --packages-above-and-dependencies tf2
TEST args: --packages-above tf2
ROS Distro: humble
Job: ci_launcher
ci_launcher ran: https://ci.ros2.org/job/ci_launcher/18548

• Linux
• Linux-aarch64
• Linux-rhel
• Windows

[ahcorde]

@selimrecep Do you mind to review backport code in Humble? it's not compiling

[selimrecep]

Issue seems to be using extrapolation_error_code variable which seems to be defined in later versions of geometry2, which seems to be fetched through ... / gather() / cache->getData() , instead of changing underlying error code detection I recommend going along with the way humble returns the error code in case of an extrapolation. Just returning tf2::TF2Error::TF2_EXTRAPOLATION_ERROR should be enough

Despite I don't have a setup for "humble" yet, I will push a small change that should fix the issue

[selimrecep]

#900 should fix this issue