Simpler, sturdier stateful time tracking in both SDKs

[teh-cmc]

(Probably easier to review commit by commit)

Follow up to my discussion with @abey79 regarding his poor experience with time tracking, which was summarized in #2501 (comment):

• When implementing the step timeline of the scalar example, I first searched for set_time_sequence (used in Python API) in the Rust docs, which I found in RecordingStream. Turns out it's not in 0.7 and bugged on main (or rather ignored by MsgSender). Again, it compiled and displayed no error, but no timeline was created.

This PR makes it so that RecordingStream is always in charge of injecting its internal clock into outgoing rows (unless the caller ask it not to, e.g. because the data is meant to be timeless).
This is pretty similar to what was already in place for log_tick, except it now applies to every timelines, whether they are builtin or user defined.

• Within the Python SDK, this gets rid of all the existing manual time injection stuff.
• On the Rust SDK's side, this fixes the issue that MsgSender used to ignore the internal clock altogether (i.e. the stateful time APIs were not supported at all for Rust users).
• And finally this cleans up the Rust examples a bunch since we now have access to stateful time.

---

PR Build Summary: https://build.rerun.io/pr/2506

Docs preview: https://rerun.io/preview/178edf5/docs
Examples preview: https://rerun.io/preview/178edf5/examples

[Wumpf]

awesome! This alone makes the Rust SDK sooo much better!

Given that we have 1f64.into() and the like everywhere it would be a nice addition to change the signature from
pub fn set_time_seconds(&self, timeline: &str, seconds: Option<f64>) {
to
pub fn set_time_seconds<T: Into<Option<f64>>(&self, timeline: &str, seconds: T) {
thoughts?
(set_nanos respectively)

Rerun getting started guide needs to be adjusted though

[Wumpf]

it looks like right now we're no longer using this timepoint anywhere as we're always seem to use the ones set on the recording (plus the ones the recording generates on adding a row)
I find it still very useful to have this here additionally (also makes logically a lot of sense), but feels like we're missing something in the (code) docs and examples about this 🤔 : MsgSender::timepoint doc string doesn't talk about this and we don't seem to have an example now that uses with_timestamp

[teh-cmc]

We're still using that timestamp the same way we did before, it gets added directly into the DataRows that MsgSender generates (MsgSender::into_rows).

The only somewhat weird part is that these timestamps get potentially overridden by RecordingStream if the internal clock already has values for the same timelines.
The reason this is weird is because MsgSender shouldn't exist anymore to begin with, it's just annoying and serves no real purpose anymore... it'll go away with codegen, in the meantime I'll mark with_timestamp & friends as #[doc(hidden) and update the starting guide.

[Wumpf]

perfect