Improve the whitening sequence

[matt-knight]

The whitening sequence implemented in gr-lora was experimentally derived by @matt-knight during the course of his initial blind signal analysis. While it has been adjusted over time, it is still riddled with errors. Additionally, the portion of the whitening sequence that applies to the PHY header (first 8 octets of the sequence, or 8 symbols at 4/8) is not correct due to limitations of the Microchip RN2903 being used to deduce the sequence. Thus, the whitening sequence should be:

• Updated for accuracy
• Extended to cover the first 8 symbols/PHY header

[matt-knight]

Complete for implicit header modes. Explicit header implementation will be tracked by #3.

For anyone interested, I added /tools/whitening_consensus_tool.py to help improve whitening sequence accuracy. One feeds it an arbitrary number of whitening sequences (dumped from the decoder) and it does bitwise comparisons among all of said sequences to find the best fit. Have a look at the #if branch in lib/decoder_impl.cc to see which decoding stages should be bypassed in this operation.