linux-xtensa-uclibc container produces instructions using little-endian encoding

[ghidra1]

The configuration for this container indicates:

CT_ARCH_BE=y
CT_ARCH_ENDIAN="big"

The Xtensa compiler produces binaries which are identified as big-endian ELF, however the instructions are using little-endian encoding. The data regions appear to use big-endian. Is this intentionally using a mixed-endian convention? Is there any way to get compiler to produce pure-big-endian or pure-little-endian (i.e., instructions and data use the same endianess setting)?