Don't overwrite already set accum_t, fix pointwise output resolution

[jmitrevs]

Description

Modify Layer._set_accum_t() to not overwrite an already existing accum_t, which can happen in an optimizer. Also, modify the pointwise init functions to do nothing, since the values are already set. Otherwise, the result_t was overwritten, potentially nullifying an explicit value set by a quantizer (a Quant node in a failing test).

Type of change

• Bug fix (non-breaking change that fixes an issue)

Tests

The failing test that this fixes will be added soon. It is important that it doesn't break anything else.

TODO: We should confirm that there are no other optimizers that may cause the same probelm

Checklist

• I have read the guidelines for contributing.
• I have commented my code, particularly in hard-to-understand areas.
• I have made corresponding changes to the documentation.
• My changes generate no new warnings.
• I have installed and run pre-commit on the files I edited or added.
• I have added tests that prove my fix is effective or that my feature works.

[jmitrevs]

This needs fastmachinelearning/example-models#19 to be merged.

[jmitrevs]

With the updated example-models, and existing test (test_qonnx.py::test_sep_conv[Vitis]) fails in main but passes with this change. For the TODO, I moved the optimizer to merge the linear activation to later, so I do not think there are any other layers that will fail in this way.