Improvements to memory management in fpga_download

[sylefeb]

This pull request contains three commits to implement the following changes:

• Exit the main download loop if an error occurs, such that the badge stays in error.
• Add a memory allocation failure check on inserting a data block in the req list.
• Allow the data buffer to be captured in the req list record so as to avoid allocating twice the size of the data block in memory.

[smunaut]

Upon looking close, why is exiting the main loop needed ? There is a 1s delay for the user to read the error before resetting, so that should be enough for the user to see something went wrong, that's what's used for the other errors.

The rest looks good to me.

[sylefeb]

The intent is to avoid the message to go unnoticed if not looking at the badge. Typically the error occurs while fpga.py is uploading the data block (in this case it was after a long upload when adding to the req list). The error is shown but disappears as fpga.py keeps going: the bitstream is uploaded next, it uploads fine and it seems all is ok.
The idea is to exit the loop to ensure the badge is stuck on the error regardless of what fpga.py sends next. An alternative would be for fpga.py to stop if it can receive an error code?