Fix tensor shapes for elementwise binary operations with broadcasting

[soumyac1999]

Description of changes:

Example: [16, 1, 2] + [16, 1552, 1] has output shape of [16, 1552, 2] but the existing code was setting the size to [16, 1, 2].

Related Issues:

Linked Issues:

• Issue #

Issues closed by this PR:

• Closes Investigate possible issues in output shape computation in keras/layers/merge.py #617

Before merging:

• Did you update the flexflow-third-party repo, if modifying any of the Cmake files, the build configs, or the submodules?

---

This change is 

[jiazhihao]

The PR lgtm. However, it seems the changes do not pass the following test

$EXE "$FF_HOME"/examples/python/keras/elementwise_mul_broadcast.py -ll:py 1 -ll:gpu "$GPUS" -ll:fsize "$FSIZE" -ll:zsize "$ZSIZE" -b ${BATCHSIZE} --only-data-parallel

[soumyac1999]

The PR lgtm. However, it seems the changes do not pass the following test

$EXE "$FF_HOME"/examples/python/keras/elementwise_mul_broadcast.py -ll:py 1 -ll:gpu "$GPUS" -ll:fsize "$FSIZE" -ll:zsize "$ZSIZE" -b ${BATCHSIZE} --only-data-parallel

@jiazhihao, similar fixes were needed in Keras to correctly compute the output shapes. This test now passes (alteast locally, waiting for the CI checks).

[soumyac1999]

Moved to #1234