mtmd : add video input support (#24269)

* wip

* ok: lazy bitmap API

* remember to free lazy text

* wip

* add mtmd_helper_video

* support video input on server (base64 input)

* add MTMD_VIDEO config

* add timestamp

* update CLI

* cli: allow auto-completion for video

* add --video arg

* fix build

* update docs

* rename as suggested

tools/mtmd/mtmd.h

@@ -128,6 +128,9 @@ MTMD_API bool mtmd_support_audio(const mtmd_context * ctx);
 // return -1 if audio is not supported
 MTMD_API int mtmd_get_audio_sample_rate(const mtmd_context * ctx);
 
+// get the current marker string
+MTMD_API const char * mtmd_get_marker(const mtmd_context * ctx);
+
 // mtmd_bitmap
 //
 // if bitmap is image:
@@ -156,6 +159,34 @@ MTMD_API void                  mtmd_bitmap_free       (mtmd_bitmap * bitmap);
 MTMD_API const char * mtmd_bitmap_get_id(const mtmd_bitmap * bitmap);
 MTMD_API void         mtmd_bitmap_set_id(mtmd_bitmap * bitmap, const char * id);
 
+// mtmd_bitmap lazy
+//
+// this is a special bitmap that:
+// - does not hold the actual data
+// - can be expanded into one or more chunks (either media to text chunks)
+// user must provide a callback to fill in the data when mtmd_tokenize() is called
+// this is useful for large video inputs:
+// - allow reading video frame by frame, without loading the entire video into memory
+// - allow tracking the whole video with a single ID (for example, the file hash)
+
+// set (*out_bitmap) to non-nullptr to emit a bitmap chunk; it will be freed automatically
+// set (*out_text) to non-nullptr to emit a text chunk; it must be heap-allocated, null-terminated and will be freed automatically
+// either out_bitmap or out_text can be set, but not both
+// out_bitmap cannot be another lazy bitmap (no nested lazy allowed)
+// return value:
+//    0 on success
+//   -1 on EOF (signal to mtmd_tokenize to move on)
+//   -2 on error (signal to mtmd_tokenize to abort)
+typedef int(* mtmd_bitmap_lazy_callback)(
+    size_t chunk_idx,
+    void * user_data,
+    mtmd_bitmap ** out_bitmap,
+    char ** out_text);
+
+MTMD_API mtmd_bitmap * mtmd_bitmap_init_lazy(mtmd_context * ctx,
+                                             const char * id, // usually set to file hash
+                                             void * user_data,
+                                             mtmd_bitmap_lazy_callback callback);
 
 // mtmd_input_chunks
 //