Characterization-based FIFO sizing for non-linear models

[fpjentzsch]

Work-in-progress

Extends the characterization-based FIFO sizing algorithm to work better with non-linear models (e.g., ResNets).
To minimize rtlsim, the goal is to rely mostly on the new analytical approach to characterize the I/O behavior.

Includes testing for FIFO tightness.

Depends on:

• Analytical FIFO sizing #1185 (is included)
• Fix incorrect value_info setting in MergeONNXModels fastmachinelearning/qonnx#152 (will need update of QONNX commit hash)