Refactor delay calculation

[osherlock1]

Summary

Refactored the code handling the matched filter delay calculation and interpolation for sub-sample delay. Additionally. developed tests for the delay calculations.

• Moved upsample(), scale_rx_signal(), calculate_sub_sample_delay_parabolic(), and calculate_sub_sample_delay_zp() from script-level definitions into src/ofdm/channel/delay.py
• calc_delay.py is now pure orchestration file I/O, argparse, and calling package functions
• Added conftest.py with shared session-scoped fixtures to eliminate repeated setup across test files
• Added test classes for all new delay functions including sub-sample accuracy tests

What changed

src/ofdm/channel/delay.py

• Added upsample() FFT-based signal upsampling
• Added scale_rx_signal() normalizes signal amplitude to 0.9 max
• Added calculate_sub_sample_delay_parabolic() sub-sample delay via parabolic interpolation
• Added calculate_sub_sample_delay_zp() sub-sample delay via zero-padded FFT interpolation
• Added _correlate() internal helper shared by both delay estimators

scripts/delay/calc_delay.py

• Removed all DSP function definitions, now imports them from the package
• Fixed bug: double .append() call in the else branch of main()
• Fixed argparse ordering parse_args() now called before loading config files

tests/conftest.py

• Added shared fixtures: cfg, active_bins, pilot_vals, pilot_symbol_freq, data_symbol_freq, pilot_symbol_time, data_symbol_time, generator

tests/test_delay.py

• Added TestUpsample, TestScaleRxSignal, TestCalculateSubSampleDelayParabolic, TestCalculateSubSampleDelayZP
• Tests cover output shape, known integer delay recovery, sub-sample delay accuracy, and edge cases
• Bug caught by tests: scale_rx_signal() raised UnboundLocalError on zero input fixed