source-bigquery-batch: Fix DATETIME serialization

BigQuery DATETIME columns only support microsecond precision, but
the default serialization of the Go `civil.DateTime` struct uses
nanosecond precision. Which is fine for output only, and *also*
is fine for cursors until the capture restarts, and *also* is
fine when the datetime values are whole seconds.

But a fractional-second datetime used as a cursor will be stored
as a string in the state checkpoint, and trying to feed that back
to BigQuery after a restart will produce an error because there
are too many digits of precision.

The fix is to handle the formatting of `civil.DateTime` values
into strings ourselves and make sure it only has six digits.

source-bigquery-batch/main.go

@@ -6,8 +6,10 @@ import (
 	"fmt"
 	"math"
 	"strings"
+	"time"
 
 	"cloud.google.com/go/bigquery"
+	"cloud.google.com/go/civil"
 	"github.com/estuary/connectors/go/schedule"
 	schemagen "github.com/estuary/connectors/go/schema-gen"
 	boilerplate "github.com/estuary/connectors/source-boilerplate"
@@ -62,8 +64,15 @@ func (c *Config) SetDefaults() {
 	}
 }
 
+const (
+	// Google Cloud DATETIME columns support microsecond precision at most
+	datetimeFormatMicros = "2006-01-02T15:04:05.000000"
+)
+
 func translateBigQueryValue(val any, fieldType bigquery.FieldType) (any, error) {
 	switch val := val.(type) {
+	case civil.DateTime:
+		return val.In(time.UTC).Format(datetimeFormatMicros), nil
 	case string:
 		if fieldType == "JSON" && json.Valid([]byte(val)) {
 			return json.RawMessage([]byte(val)), nil