filter/ffmpeg: support double-precision samples

Insert an "aformat" filter which converts double-precision to
single-precision.

Closes https://github.com/MusicPlayerDaemon/MPD/issues/1235

src/lib/ffmpeg/Filter.cxx

@@ -130,6 +130,14 @@ MakeAformat(AudioFormat &audio_format,
 			    args, nullptr, graph_ctx);
 }
 
+AVFilterContext &
+MakeAutoAformat(AVFilterGraph &graph_ctx)
+{
+	return CreateFilter(RequireFilterByName("aformat"), "aformat",
+			    "sample_fmts=flt|s32|s16",
+			    nullptr, graph_ctx);
+}
+
 void
 FilterGraph::ParseSingleInOut(const char *filters, AVFilterContext &in,
 			      AVFilterContext &out)

src/lib/ffmpeg/Filter.hxx

@@ -103,6 +103,13 @@ AVFilterContext &
 MakeAformat(AudioFormat &audio_format,
 	    AVFilterGraph &graph_ctx);
 
+/**
+ * Create an "aformat" filter which automatically converts the output
+ * to a format supported by MPD.
+ */
+AVFilterContext &
+MakeAutoAformat(AVFilterGraph &graph_ctx);
+
 class FilterGraph {
 	AVFilterGraph *graph = nullptr;