readme : document the ANEForge encoder backend (#4073)

* readme : document the ANEForge encoder backend

* Improve clarity in ANEForge support section of README

Reworded sentences for clarity and improved readability in the ANEForge section.

* Improve formatting of ANEForge support section

Reformat ANEForge section for better readability.

README.md

@@ -232,6 +232,31 @@ speed-up - more than x3 faster compared with CPU-only execution. Here are the in
 
 For more information about the Core ML implementation please refer to PR [#566](https://github.com/ggml-org/whisper.cpp/pull/566).
 
+## ANEForge support
+
+On Apple Silicon, the Encoder can also run on the Apple Neural Engine via [ANEForge](https://github.com/sbryngelson/ANEForge), which dispatches to the ANE directly instead of through Core ML.
+It is about 2x faster than the Core ML encoder from `tiny` to `medium` ([benchmarks](https://github.com/sbryngelson/ANEForge/tree/main/bench/whisper_encoder_ane)).
+The Decoder is unchanged, and no build flag is needed.
+
+Compile the encoder into a bundle (tied to the machine and OS build that produced it):
+
+```bash
+git clone https://github.com/sbryngelson/ANEForge && cd ANEForge
+pip install -e ".[models]"
+PYTHONPATH=. python3 bench/whisper_encoder_ane/export_bundle.py \
+    --model openai/whisper-base --out /tmp/whisper-base-encoder
+```
+
+Then point `whisper.cpp` at it:
+
+```bash
+export ANEFORGE_ENCODER=/tmp/whisper-base-encoder
+export ANEFORGE_DYLIB=$PWD/aneforge/_lib/libane_e5rt_dispatch.dylib
+./build/bin/whisper-cli -m models/ggml-base.bin -f samples/jfk.wav
+```
+
+For more information about the ANEForge implementation, see PR [#3905](https://github.com/ggml-org/whisper.cpp/pull/3905).
+
 ## OpenVINO support
 
 On platforms that support [OpenVINO](https://github.com/openvinotoolkit/openvino), the Encoder inference can be executed