This change refactors the Neural Residual Echo Estimator (ML REE) to properly handle multi-channel audio signals for all inputs.

  All multi-channel handling, buffering, and pre-processing is moved to the FeatureExtractor classes.

  Key changes:
   - The main NeuralResidualEchoEstimator::Estimate interface is updated to accept a multi-channel webrtc::Block for the render signal.
   - NeuralResidualEchoEstimatorImpl no longer performs any buffering or channel summation.
   - The FeatureExtractor interface is redesigned, splitting responsibilities into:
     - UpdateBuffers: For buffering incoming blocks of data.
     - PrepareModelInput: For preparing the model's input tensor when enough data is available.
     - ReadyForInference: To check if a full frame can be processed.
     - TimeDomainFeatureExtractor now implements channel summation for all inputs.
   - FrequencyDomainFeatureExtractor buffers each channel and then computes the average power spectrum across all channels.

Bug: webrtc:442444736
Change-Id: Ie0df331c87a414146187f661aed17369a946d660
No-Iwyu: Not changing the tflite includes.
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/429142
Reviewed-by: Sam Zackrisson <saza@webrtc.org>
Commit-Queue: Jesus de Vicente Pena <devicentepena@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#46345}