Store dateTime statements as Neo4j datetime values

[malberts]

For #831

Statement values for the dateTime property type were written to Neo4j as
plain ISO 8601 strings, leaving them un-queryable with Cypher's datetime
functions (comparisons, ordering, range filtering).

Value builders registered with Neo4jValueBuilderRegistry can now return
typed objects: the Neo4j driver converts DateTimeInterface instances
nested in the props map to native Bolt temporal values, so the single-map
SET n = $props write stays intact and no Cypher constructor expressions
or query-string assembly are needed. The dateTime builder parses statement
strings with DateTimeProperty::parseStrictDateTime(); malformed values
are omitted from the graph projection (the revision slot stays
authoritative). Extension-defined property types get the same capability
by returning typed objects (e.g. Laudis Date, Duration, points) from
their builders.

The result normalizer converts temporal values in query results to
ISO 8601 strings, and other Cypher property objects to arrays, instead of
throwing Unsupported Cypher value type (previously reachable via
RETURN datetime()).

Existing graph data is re-typed by the standard rebuild
(maintenance/RebuildGraphDatabases.php); until then, pre-existing
string-stored values silently don't match temporal comparisons.

The page-side typed path (PageValue / Neo4jQueryStore::extractTypedValues())
is deliberately left untouched.

Follow-ups (out of scope, filed separately)

• The date property type still stores plain strings (same un-queryability this
  fixes for dateTime): Store date statements as native Neo4j dates so date comparisons work #989
• The page-facts datetime write path emits zone-less strings that shift under a
  non-UTC Neo4j timezone: Page-facts datetime path stores zone-less strings, shifting under a non-UTC Neo4j timezone #990

First iteration (Cypher constructor wrapping) written by Claude Code
Opus 4.7 (1M context) with @malberts. Reworked to driver-native typed
parameters with @JeroenDeDauw, following an adversarial multi-agent
design review of #831 against the codebase and the Laudis client.
_{Written by Claude Code, Fable 5 (max)}

[JeroenDeDauw]

@malberts ready?

[malberts]

@JeroenDeDauw Ready after #1004. There are some minor follow-ups to do here, which CC got too excited to post already.