F/tflite parser

[neil-tan]

uTensor frontend that imports .tflite to ugraph.
Main file: utensor_cgen/frontend/tflite.py

Currently supported op:

QUANTIZE
DEPTHWISE_CONV_2D
MAX_POOL_2D
RESHAPE
FULLY_CONNECTED
DEQUANTIZE
ARG_MAX

The code-snippet for training and exporting the .tflite file:

converter = tf.lite.TFLiteConverter.from_saved_model('saved_model/test_model')
converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.representative_dataset = representative_dataset_gen
tflite_model = converter.convert()

Test command:
utensor-cli convert converted_model.tflite

.toml modification:
transform_methods = [ "dropout(name_pattern=r'(dropout[_\\w\\d]*)/.*')", "graph_viz", ]

The test .tflite file will be kept available here until the PR is merged.

[dboyliao]

1. Please remove unused imports.
2. There are return type inconsistency in your op options parsing function and it should not work.
3. Please provide at least one test for your parser
4. The model file downloadable in your PR description looks like a normal model instead of a quantized one. That's the reason why the parser works on that file since both quantize_op_data and dequantize_op_data are not called. Please double check the model file.

[dboyliao]

BTW, the op's names look weird.

[dboyliao]

@neil-tan
I fix the b' prefix issue.
However, I'd like you to remove the :0 from the op names.

[dboyliao]

Looks good to me.
Merge