Avoid keyword args to njit functions

[tornikeo]

Describe the bug
For some odd reason, adding default values to keyword args to @njit-decorated functions results in 20x slowdowns in frequently used functions, like collect_peak_pairs. Further optimizations can bring this particular function to 40x speedup.

To Reproduce
Steps to reproduce the behavio:

• Just follow this colab notebook.

It shows that original collect_peak_pairs takes around

63.5 µs ± 1.06 µs per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

While optimized takes around:

1.52 µs ± 26.9 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

What else?
Given that this function is the innermost segment in a double python for-loop, I don't know how much of a total runtime impact this will have - that outer loop has to be optimized too to take full advantage of this.