[stdlib] Implement Python-like base64 encoding/decoding

[hayduker]

This PR implements a few features to make Mojo's base64 encoding and decoding conform more closely to Python's implementation:

• b64decode now ignores whitespace in the input, fixing [BUG] b64decode does not handle whitespaces #3446.
• Both b64encode and b64decode implement an optional altchars argument which allows the caller to substitute two characters for the default +/, useful for URL-safe encoding.
• b64decode implements an optional validate argument which, when True, will cause the function to raise an Error if the input contains any characters outside the alphabet. When False (the default), these characters are ignored like whitespace.
• b64decode now raises an Error if the input's length (after whitespace / invalid character removal) isn't divisible by 4.

Note about my comment about "leaky abstraction" in b64encode: I tried using String.unsafe_ptr() instead of String._buffer directly, since this was the function used for StringLiteral originally, but it does not return the correct underlying UInts corresponding to the characters in the string. I ran into this issue switching from using a StringLiteral alias to String variable for the alphabet, since I need to modify the alphabet according to the altchars argument.

I'm not sure if this is a bug or just my misunderstanding UnsafePointer. Here is a small example that exhibits what I mean:

def main():
    var l = "ABC"
    var lp = l.unsafe_ptr()
    print(lp[0])  # returns 65

    var s = String("ABC")
    var sp = s.unsafe_ptr()
    print(sp[0])  # returns 0

If this is a bug and I should open an issue for it, let me know.

[soraros]

FYI, there is already #3443.

[hayduker]

Oh cool, not sure how I missed that. Closing this.