ContentDateEnrichment: Filter _update_by_query to only unresolved documents (#3118)

* ContentDateEnrichment: Filter _update_by_query to only unresolved documents

The _update_by_query in ResolveContentDatesAsync was re-indexing every
document in both the lexical and semantic indices. On the semantic index,
this triggered ML inference for all 6 semantic_text fields on every
document — causing the deploy workflow to hang for 3+ hours.

After HashedBulkUpdate, unchanged documents (noop) retain their resolved
content_last_updated from the previous run. Only new/changed documents
have the field at the default DateTimeOffset.MinValue (0001-01-01). The
filter restricts _update_by_query to only these unresolved documents,
reducing the typical deploy from hundreds of thousands of documents to
just the changed ones.

Also enhances integration tests to use real HashedBulkUpdate-style
scripted upserts with full DocumentationDocument serialization, and adds
tests proving the filter behavior.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Simplify: Clean up review findings

- Extract query JsonObject to static readonly string (avoid re-allocation per call)
- Remove debug output.WriteLine from discovery test
- Fix double JsonNode.Parse — reuse parsed node for params.doc
- Use const for hash field name
- Fix misleading comment on doc1 in FilteredResolve test
- Remove unnecessary null-forgiving operators

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Tests: Assert bulk response has no item-level errors

The existing check only verified HTTP status code, which can be 200 even
when individual bulk items fail. Parse the response body and assert
"errors": false.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Tests: Fail explicitly on missing or malformed bulk errors field

Assert bulkResult and its "errors" property exist before checking the
boolean value, rather than defaulting to false on missing/malformed JSON.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: reakaleek

src/Elastic.Markdown/Exporters/Elasticsearch/ContentDateEnrichment.cs

@@ -77,11 +77,35 @@ public async Task SyncLookupIndexAsync(string lexicalAlias, Cancel ct)
 	/// the pipeline compares each document's content_hash against the lookup from
 	/// the previous run and either preserves the old date or stamps a new one.
 	/// </summary>
+	// Only process documents that don't already have a valid content_last_updated.
+	// After HashedBulkUpdate: unchanged docs (noop) retain their resolved date from the
+	// previous run; new/changed docs have the field missing or at DateTimeOffset.MinValue
+	// (0001-01-01). This filter avoids re-indexing the entire index, which on the semantic
+	// index would trigger expensive semantic_text inference for every document.
+	private static readonly string UnresolvedContentDatesQuery = new JsonObject
+	{
+		["query"] = new JsonObject
+		{
+			["bool"] = new JsonObject
+			{
+				["must_not"] = new JsonArray(
+					new JsonObject
+					{
+						["range"] = new JsonObject
+						{
+							["content_last_updated"] = new JsonObject { ["gt"] = "1970-01-01T00:00:00Z" }
+						}
+					}
+				)
+			}
+		}
+	}.ToJsonString();
+
 	public async Task ResolveContentDatesAsync(string indexAlias, Cancel ct)
 	{
 		logger.LogInformation("Resolving content dates in {Index} via pipeline {Pipeline}", indexAlias, PipelineName);
 
-		await operations.UpdateByQueryAsync(indexAlias, PostData.Empty, PipelineName, ct);
+		await operations.UpdateByQueryAsync(indexAlias, PostData.String(UnresolvedContentDatesQuery), PipelineName, ct);
 
 		logger.LogInformation("Content date resolution complete for {Index}", indexAlias);
 	}