encrypt_key and decrypt_key were using a feature of sum() where
the second argument is a 'start' value and every futher value
is += onto that. Since it was a tuple, the final result is a tuple.
Replaced it with a tuple comprehension.
get_chunks just needed some variable names but I am not sure the
purpose of the increasing chunk size during iteration. It's probably
a feature of the MEGA webclient.
I noticed that _init_shared_keys assigns the result to the instance,
so let's just keep it on the instance from the beginning and not
pass it around. If some individual keys get overwritten during runtime
I expect that would be ok.
This makes the caller's life easier. They can catch
mega.errors.EFAILED instead of catching mega.errors.RequestError
and checking the message attribute.
Because these classes inherit from RequestError, and they have a code
and message attribute, anyone currently catching RequestError should
not have any backwards compatibility issues.
Furthermore, this fixes an existing issue in the codebase where
RequestError is raised with a custom string message, which was causing
IndexError since that message wasn't in CODE_TO_DESCRIPTIONS.
The code was using a hardcoded 257 as the RSA public exponent, but
it was raising RSA Invalid Construct. From reading MEGA's webclient js
I found that the public exponent sometimes defaults to 257, but in
other cases is calculated from a modular inverse on the private
exponent and phi=p-1*q-1.
Updated the reqs to use pycryptodome, fixed the RSA public exponent issue that was causing the logn failure, as well as general updates to patch up slight changes.
