Add flag to APM to force multichannel even with AEC3

Currently, APM fakes multichannel in two ways:
 - With injected AECs, capture processing is only performed on the left
channel. The result is copied into the other channels.
 - With multichannel render audio, all channels are mixed into one
before analysing.

This CL adds a flag to disable these behaviors, ensuring proper
multichannel processing happens throughout the APM pipeline.

Adds killswitches to separately disable render / capture multichannel.

Additionally - AEC3 currently crashes when running with multichannel.
This CL adds the missing pieces to at least have it run without
triggering any DCHECKS, including making the high pass filter properly
handle multichannel.

Bug: webrtc:10913, webrtc:10907
Change-Id: I38795bf8f312b959fcc816a056fba2c68d4e424d
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/152483
Commit-Queue: Sam Zackrisson <saza@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#29248}

modules/audio_processing/include/audio_processing.h

@@ -254,6 +254,9 @@ class AudioProcessing : public rtc::RefCountInterface {
       // default rate is currently selected based on the CPU architecture, but
       // that logic may change.
       int maximum_internal_processing_rate;
+      // Force multi-channel processing on playout and capture audio. This is an
+      // experimental feature, and is likely to change without warning.
+      bool experimental_multi_channel = false;
     } pipeline;
 
     // Enabled the pre-amplifier. It amplifies the capture signal