better version of load_endpoints func

[kqiq]

I moved the update of the headers dictionary to the beginning of the function, to avoid creating a new dictionary on each iteration of the loop. I replaced the ext_bin_headers dictionary comprehension with a regular loop and an update to the headers dictionary, which should be faster than creating a new dictionary. I replaced the ** operator with the update() method to merge the dictionaries, which is generally faster than using the ** operator.