Merge branch 'master' into v4

Resolve 57 conflicts keeping v4's architecture (UI removal, api app
context, path isolation, streaming/codecs/rtc, workflow-mode superset,
args_helper) and taking master's post-fork features so none are dropped:
inference override/adjust provider hooks (+ face_swapper 3.8.1 coreml
fix), is_vision_frame validity guards, resolve_temp_frame_set frame
numbering with the source audio/voice trim offset, the ffmpeg color
pipeline (restrict_color_transfer / convert_color_space / temp_pixel_format),
ffprobe select_stream v:0 + format duration, pre_check ffprobe with
dependency_not_installed, onnxruntime arena-leak guard + version lru_cache,
installer cuda@12/@13 provider split, workflow_strategy disk/memory
(memory default), and the dependency version bumps. content_analyser hash
guard rehashed to 3c6ce25e.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V7MdZEmd1GE8uSDTMyq34r

facefusion.ini

@@ -1,6 +1,3 @@
-[workflow]
-workflow_mode =
-
 [paths]
 temp_path =
 jobs_path =
@@ -54,6 +51,7 @@ voice_extractor_model =
 trim_frame_start =
 trim_frame_end =
 temp_frame_format =
+temp_pixel_format =
 
 [frame_distribution]
 target_frame_amount =
@@ -71,6 +69,10 @@ output_video_quality =
 output_video_scale =
 output_video_fps =
 
+[workflow]
+workflow_mode =
+workflow_strategy =
+
 [processors]
 processors =
 age_modifier_model =