Memory: add multimodal image and audio indexing (#43460)

Merged via squash.

Prepared head SHA: a994c07190
Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com>
Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com>
Reviewed-by: @gumadeiras

src/memory/embedding-inputs.ts

@@ -0,0 +1,34 @@
+export type EmbeddingInputTextPart = {
+  type: "text";
+  text: string;
+};
+
+export type EmbeddingInputInlineDataPart = {
+  type: "inline-data";
+  mimeType: string;
+  data: string;
+};
+
+export type EmbeddingInputPart = EmbeddingInputTextPart | EmbeddingInputInlineDataPart;
+
+export type EmbeddingInput = {
+  text: string;
+  parts?: EmbeddingInputPart[];
+};
+
+export function buildTextEmbeddingInput(text: string): EmbeddingInput {
+  return { text };
+}
+
+export function isInlineDataEmbeddingInputPart(
+  part: EmbeddingInputPart,
+): part is EmbeddingInputInlineDataPart {
+  return part.type === "inline-data";
+}
+
+export function hasNonTextEmbeddingParts(input: EmbeddingInput | undefined): boolean {
+  if (!input?.parts?.length) {
+    return false;
+  }
+  return input.parts.some((part) => isInlineDataEmbeddingInputPart(part));
+}