1594 twist stamped publisher

[Ryanf55]

See #3775; this is the same PR but opened back up.

Replaces #4031, by pulling in all 3 commits, as well as more changes since then.

This has been tested by backporting to humble against ArduPilot by @srmainwaring. No bugs were reported.

[SteveMacenski]

Note to self once ripping out non-stamped:

• Check constructor for stamped only is OK
• Change all uses of publisher / subscriber to the stamped only option
• Delete all non-stamped function overrides
• Re-check all timestamps and headers are properly propagated (not reassigned, overrided, etc after time of computation of the root cmd).
• Remove enabled stamped param from param guide + mention in migration guide total deprecation
• Review carefully for what I might have missed now

[Ryanf55]

I spent a bit of time looking into how ABI broke, and whether it's possible to preserve it. This is purely informative for the situation that someone requests a backport to humble or iron.

Steve and I discussed this over Slack, and the plan for now, although different than I originally planned, is to not do a backport, and instead just plan on having this be part of ROS 2 Jazzy. ArduPilot users on 4.5 (released yesterday) will not be able to use NAV2 for controlling a vehicle until all of ArduPilot's code is upgraded to run on Jazzy.

An ABI-breaking backport of this PR to humble, for those that need it, is here:
https://github.com/Ryanf55/navigation2/tree/backport-humble-1594
Rhys and I have tested this in SITL, and it works. The backport was quite messy with merge conflicts, so I don't plan to maintain this forked branch.

If a backport is requested, Steve recommends that

1. ABI breaks won't be allowed so the PR must be modified before merge.
2. We must allow enough time with this on rolling to make sure it is working well before we backport to a stable distro

This may be worth it, as it's going to be a while before you can run ROS 2 Jazzy on Nvidia hardware without docker. Jetpack 6 releases this spring for LTS support on Jetson Orin. Without a backport, it will force users of NAV2 to continue to rely on docker to get a newer OS version than is supplied by common vendors like Nvidia.

Another potential option is to just add Twist support in ArduPilot, and mark it deprecated from the get-go.

Here is my list of the primary ABI break violations and potential strategies.

Hiding private members

Looks like at least one ABI break is due to non-compliance with google style guide:
https://github.com/ros-planning/navigation2/pull/4077/files#diff-aefeea6b3229aeefd5f824e1dad301a559b6cf016bb47e32491943300d63f49bL103

Data members are supposed to be private, however in this class, it's protected, thus it's part of ABI when it's not clear why.
https://google.github.io/styleguide/cppguide.html#Access_Control

There is not a good path to not break ABI in this scenario other than storing both data members which could be confusing and buggy.

Passing raw pointers

This ABI break is due to passing a raw pointer and me changing that.
https://github.com/ros-planning/navigation2/pull/4077/files#diff-a17c9ff7afe40ec85e895af17b5565c69ad128754d4a4994b54b827c9e7f503cL165

Instead of a raw pointer, smart pointers are recommended.
https://google.github.io/styleguide/cppguide.html#Ownership_and_Smart_Pointers

This change could be reverted, and an override added instead.

Callbacks being exposed protected instead of private

Another ABI break is in relation to callbacks being changed.

Here, a protected callback was removed and put inline:
https://github.com/ros-planning/navigation2/pull/4077/files#diff-aefeea6b3229aeefd5f824e1dad301a559b6cf016bb47e32491943300d63f49bL89

This callback is only used internally right now, but since it was exposed publicly, it would need to be preserved, even if internally.

Const pointers

Here, a pointer was passed non-const, even though it was not intended to be modified.
https://github.com/ros-planning/navigation2/pull/4077/files#diff-a17c9ff7afe40ec85e895af17b5565c69ad128754d4a4994b54b827c9e7f503cL165

Google style strongly recommends using const when it makes sense.
https://google.github.io/styleguide/cppguide.html#Use_of_const

Overrides could be added instead of changing the original method.

Addition of header

In this change, a header is added to the member function:
https://github.com/ros-planning/navigation2/pull/4077/files#diff-1f10ee4d90224cdf0df5406c16550451120bdf02367ca835551eb99a2b061faaL104

To preserve ABI, one could make this second argument optional, or add an overload.

Addition of nodiscard

Here, nodiscard was added to return values: https://github.com/ros-planning/navigation2/pull/4077/files#diff-eda21c4165d798f8cfac221df11dc7aed7cfac3238d317972350347c30a2b07eL106

While this protects consumers of the code, adding nodiscard breaks ABI. This change could simply be dropped, and only enforced in rolling and jazzy.