spectrum: further speed up 3GPP channel

After hours of vibecoding and testing, we get another decent speedup with almost zero impact to any of the results (channel matrix and PSD). Impact is due to associative math reordering and binary floating point rounding.

With the synthetic benchmark built by Opus, in the four config and four scenarios I asked it to evaluate we had (in microseconds per link per realization/update)

After the optimizations, we got the following speedup for GetNewChannel/DoCalcRxPower something something.

And because I'm not crazy, I tested with NR's outdoor calibration job. Exact same E2E performance in terms of throughput and latency, but more than 2x speedup.