Faster continuous k-mer decomposition

[padix-key]

This PR introduces a faster method for create_kmers() that uses the k-mer from the previous position to compute the next position. This way lopping over k is not necessary anymore as opposed to the current method.
The performance increase is clear (orange: naive, green: new method):

The following script was used to create the benchmarks:

import time
import numpy as np
import biotite
import biotite.sequence as seq
import biotite.sequence.align as align
import matplotlib.pyplot as plt


LENGTH = 1_000_000
MAX_K = 12
N = 100


orig_durations = []
new_durations = []

sequence = seq.NucleotideSequence(ambiguous=False)
np.random.seed(0)
sequence.code = np.random.randint(len(sequence.alphabet), size=LENGTH)

k_values = np.arange(2, MAX_K+1)
for k in k_values:
    kmer_alph = align.KmerAlphabet(sequence.alphabet, k)

    now = time.time_ns()
    for _  in range(N):
        kmer_codes = kmer_alph.create_kmers(sequence.code)
    new_durations.append((time.time_ns() - now) / N * 1e-6)

    kmer_alph._create_kmers_func = kmer_alph._create_continuous_kmers

    now = time.time_ns()
    for _  in range(N):
        kmer_codes = kmer_alph.create_kmers(sequence.code)
    orig_durations.append((time.time_ns() - now) / N * 1e-6)

fig, ax = plt.subplots()
ax.plot(
    k_values, orig_durations, color=biotite.colors["orange"],
    marker="o", linestyle="None"
)
ax.plot(
    k_values, new_durations, color=biotite.colors["green"],
    marker="o", linestyle="None"
)
ax.set_xlabel("k")
ax.set_ylabel("Duration (ms)")
plt.show()

For easier implementation the k-mer ordering in the alphabet is reversed.