Update of down sampling, partial hanning taper, rescale ccf, and allow device with mps

[frostforestcdjl]

1. read_mseed can get --sampling_rate and do downsample now
2. do demean before pad zeros to get one day length, or the padded zeros will be shifted after the later demean.
3. comment nt=8640001, or the tensor from "data = np.zeros([3, nx, nt], dtype=np.float32)" will create data length before downsampling instead of the data length after downsampling.
4. apply "partial_hann_taper" to replace the hann_window.
5. ccf îs devided by window length after cross-correlation now.
6. In run.py, add os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1" to allow --device=mps.