#15713: Use ZEROACC properly for FP32 case

[ncvetkovicTT]

Corrected ZEROACC call for fp32 bcasted eltwise mul. Tested by tt-metal post-commit:
https://github.com/tenstorrent/tt-metal/actions/runs/12354236372
and BH post-commit:
https://github.com/tenstorrent/tt-metal/actions/runs/12355477287

(later post-commit runs fail because there are issues with the pipeline #22539, #22474, #2408)

This PR resolves tenstorrent/tt-metal#15713

[ncvetkovicTT]

@nvelickovicTT Pipeline is broken, but I have run broadcast unit tests locally and within the two passing workflow runs (before they broke):

[       OK ] ComputeSingleTileBroadcast/BroadcastParameterizedDeviceFixture.TensixComputeSingleTileBroadcast/35 (910 ms)
[----------] 36 tests from ComputeSingleTileBroadcast/BroadcastParameterizedDeviceFixture (30752 ms total)

[----------] Global test environment tear-down
[==========] 36 tests from 1 test suite ran. (30752 ms total)
[  PASSED  ] 33 tests.
[  SKIPPED ] 3 tests, listed below:
[  SKIPPED ] ComputeSingleTileBroadcast/BroadcastParameterizedDeviceFixture.TensixComputeSingleTileBroadcast/12
[  SKIPPED ] ComputeSingleTileBroadcast/BroadcastParameterizedDeviceFixture.TensixComputeSingleTileBroadcast/21
[  SKIPPED ] ComputeSingleTileBroadcast/BroadcastParameterizedDeviceFixture.TensixComputeSingleTileBroadcast/30
                 Device | INFO     | Closing user mode device drivers