Get rid of joining indices for equal axes at pandas backend

[dchigarev]

Currently, when doing binary operations and concat we're joining indices of left and right operands, however if axes are equals we can avoid joining and that will give us decent speedup:

import pandas
import numpy as np
from timeit import default_timer as timer

N = 50_000_000

ind1 = pandas.Index(np.arange(N))
ind2 = pandas.Index(np.arange(N))

t1 = timer()
if not ind1.equals(ind2):
    result = ind1.join(ind2, how="outer")
else:
    result = ind1
print("Checking for equality:", timer() - t1)  # ~ 0.1s

t1 = timer()
result = ind1.join(ind2, how="outer")
print("Do not checking for equality:", timer() - t1)  # ~ 8.3s

[anmyachev]

Fixed in #2378