feat(core): add _ts system column

[leoyvens]

Addresses #2048.

This PR implements a _ts built-in and auto-propagated timestamp column, just like we do for _block_num. The two columns are abstracted by WatermarkColumn.

The largest concern with this PR is backwards compatibility, to not cause immediate schema mismatch errors on existing deployments. So we make sure the _ts column is only included when it is available at the source and requested by the consumer of the query.

Remaining piece of #2048 is implementing a temporary hack to recognize the timestamp EVM column as equivalent to _ts, so we don't have to resync raw datasets to use the feature.

[Theodus]

LGTM, just some suggestions.

[Theodus]

Suggested change

	if rows
	.schema()
	.column_with_name(RESERVED_TS_COLUMN_NAME)
	.is_some()
	{
	let ts_col = rows
	.column_by_name(RESERVED_TS_COLUMN_NAME)
	.ok_or(CheckInvariantsError::MissingTsColumn)?;
	if !matches!(
	ts_col.data_type(),
	DataType::Timestamp(TimeUnit::Nanosecond, Some(tz)) if tz.as_ref() == "+00:00"
	) {
	return Err(CheckInvariantsError::InvalidTsColumnType);
	}
	}
	if let Some(ts_col) = rows.column_by_name(RESERVED_TS_COLUMN_NAME)
	&& !matches!(
	ts_col.data_type(),
	DataType::Timestamp(TimeUnit::Nanosecond, Some(tz)) if tz.as_ref() == "+00:00"
	)
	{
	return Err(CheckInvariantsError::InvalidTsColumnType);
	}

[Theodus]

This error variant is dead code if the above suggestion is applied.

Suggested change

	/// Required `_ts` column is missing from the record batch
	#[error("missing _ts column")]
	MissingTsColumn,

[Theodus]

Looks like we've added these _generated.json files without removing the old ones. Is this intentional?

[leoyvens]

The situation is:

• We have tests for the output of ampctl gen and we want new manifests generated with the _ts column.
• I still wanted to keep all tests on old manifests to be sure this would not break existing datasets that do not have _ts.

So we needed two manifests, at least for now.

[Theodus]

There is an inconsistency here between the name SystemColumnPropagator and WatermarkColumnPropagator. Might be worth doing a pass to consolidate on one term (system column vs watermark column).

[Theodus]

We should probably copy after the filter

Suggested change

	.iter()
	.copied()
	.filter(|wm| {
	let name = wm.column_name();
	schemas
	.iter()
	.all(|s| s.fields().iter().any(|f| f.name() == name))
	})
	.collect()
	.iter()
	.filter(|wm| {
	let name = wm.column_name();
	schemas
	.iter()
	.all(|s| s.fields().iter().any(|f| f.name() == name))
	})
	.copied()
	.collect()

[leoyvens]

@Theodus thanks for the review, I've addressed your comments