This pull request adds support for custom input formats in the ESP_SR speech recognition library, allowing users to specify the channel layout for microphone and playback data. It also introduces a new partition scheme for ESP_SR models, updates the example and CI configuration to use these changes, and refines the build conditions for ESP_SR to use model storage options instead of wake word/multinet flags.

Speech Recognition Input Format Support:

• Added an input_format parameter to the ESP_SR_Class::begin method and the underlying sr_start function, allowing users to specify the channel layout for speech recognition (e.g., "MMNR" for two microphones, one unused, one playback channel). This is reflected in both the header and implementation files (ESP_SR.h, ESP_SR.cpp, esp32-hal-sr.h, esp32-hal-sr.c). [1] [2] [3] [4] [5]
• Updated the Basic example (Basic.ino) to use the new SR_INPUT_FORMAT define and to pass it to ESP_SR.begin. Added documentation for the input format. [1] [2]

Partition Scheme and CI Improvements:

• Added a new partition scheme esp_sr_16 for ESP32P4 boards, supporting 3MB APP, 7MB SPIFFS, and 2.9MB MODEL, with corresponding upload size and flags in boards.txt.
• Updated CI configuration (ci.json) to test ESP_SR on ESP32P4 boards with the new partition scheme. [1] [2]

Build Condition Refinements:

• Changed conditional compilation flags throughout ESP_SR source files to use CONFIG_MODEL_IN_FLASH or CONFIG_MODEL_IN_SDCARD instead of CONFIG_USE_WAKENET or CONFIG_USE_MULTINET, ensuring the code is built only when model storage is available. [1] [2] [3] [4]