Perf: optimize function pow in module_gint

[dzzz2001]

Background

When profiling the abacus program with vtune, I found that the pow function consumed a significant portion of the cal_dpsir_ylm function's runtime. std::pow is an inefficient function, especially when the exponent is a small integer, as direct multiplication can be dozens of times faster than calling std::pow. Thus, I overloaded the pow function for cases where the exponent is a small integer, which has accelerated the computation of cal_gint_force.
Before optimizing, the flame graph of cal_dpsir_ylm:

After optimizing, the flame graph of cal_dpsir_ylm(the "_INTERN..." part represents pow function):

It's clear that the time portion spent in pow function dropped significantly after the overloaded of pow.
Here is the performance comparison (the testing example is tests/performance/P103_Si128_lcao and the testing command is "OMP_NUM_THREADS=12 mpirun -n 1 abacus"):

	before optimization	after optimization
cal_gint_force(CPU)	16.81s	14.06s
cal_gint_force(GPU)	3.91s	3.01s