[Fix] Correct SNR calculation logic

[JMZ-kk]

Summary

This PR fixes an issue where the measured SNR values did not match the design targets and varied inconsistently across different MIMO configurations.

Specifically, it addresses the incorrect scaling of noise generation regarding Array Gain (total received signal power scaling with users) and Pilot Sparsity (frequency-orthogonal mode).

Problem Description

Previously, the noise generation logic did not account for the signal power scaling inherent in different MIMO configurations. For example, in files/config/examples/rp-sim-ul.json, a setting of "noise_level": 0.01 (targeted for ~40 dB SNR) resulted in inconsistent actual SNRs:

Configuration	Uplink Data SNR	Pilot SNR
64 UEs × 64 BS	53.3 dB	35.2 dB
16 UEs × 64 BS	47.1 dB	29.1 dB

Root Causes identified:

1. Array Gain: Total received signal power scales with the number of users ($N_t$), but the noise generation was constant. This caused larger MIMO setups (e.g., 64x64) to artificially have better BER than smaller ones (e.g., 16x16), which is theoretically incorrect for our test cases.
2. Pilot Sparsity: In frequency-orthogonal mode, pilots use fewer subcarriers. The previous calculation treated them as full-band. Consequently, the pilot symbol power was only $1 / \text{PilotScGroupSize}$ of the data symbol power, leading to severe SNR degradation on pilots and incorrect channel estimation ($H$).

Changes

We adjusted the noise generation to maintain a constant SNR regardless of the configuration:

• Data Symbols: Scaled noise amplitude by $\sqrt{N_t}$ to offset array gain.
• Pilot Symbols: Scaled pilot noise amplitude by $1 / \sqrt{\text{PilotScGroupSize}}$ to account for sparsity.