esp32: Enable automatic Python heap growth.

Via MICROPY_GC_SPLIT_HEAP_AUTO feature flag added in previous commit.

Tested on ESP32 GENERIC_SPIRAM and GENERIC_S3 configurations, with some
worst-case allocation patterns and the standard test suite.

This work was funded through GitHub Sponsors.

Signed-off-by: Angus Gratton <angus@redyak.com.au>

docs/library/esp32.rst

@@ -51,13 +51,20 @@ Functions
     buffers and other data. This data is useful to get a sense of how much memory
     is available to ESP-IDF and the networking stack in particular. It may shed
     some light on situations where ESP-IDF operations fail due to allocation failures.
-    The information returned is *not* useful to troubleshoot Python allocation failures,
-    use `micropython.mem_info()` instead.
 
     The capabilities parameter corresponds to ESP-IDF's ``MALLOC_CAP_XXX`` values but the
     two most useful ones are predefined as `esp32.HEAP_DATA` for data heap regions and
     `esp32.HEAP_EXEC` for executable regions as used by the native code emitter.
 
+    Free IDF heap memory in the `esp32.HEAP_DATA` region is available to be
+    automatically added to the MicroPython heap to prevent a MicroPython
+    allocation from failing. However, the information returned here is otherwise
+    *not* useful to troubleshoot Python allocation failures, use
+    `micropython.mem_info()` instead. The "max new split" value in
+    `micropython.mem_info()` output corresponds to the largest free block of
+    ESP-IDF heap that could be automatically added on demand to the MicroPython
+    heap.
+
     The return value is a list of 4-tuples, where each 4-tuple corresponds to one heap
     and contains: the total bytes, the free bytes, the largest free block, and
     the minimum free seen over time.