Improve Base64.urlsafe_encode64 performance

[jcmfernandes]

Creating this here as asked in ruby/ruby#4802

Improves the method's performance when asked to remove padding.

str.delete!("=") iterates over the entire string looking for the equals
character, but we know that we will, at most, find two at the end of the
string.

[jcmfernandes]

Some benchmarks:

Input generated with SecureRandom.alphanumeric.
.urlsafe_encode64 being called with padding: false.

4 kB input
Warming up --------------------------------------
NEW #urlsafe_encode64
14.230k i/100ms
OLD #urlsafe_encode64
7.588k i/100ms
Calculating -------------------------------------
NEW #urlsafe_encode64
141.104k (± 2.5%) i/s - 1.423M in 10.091936s
OLD #urlsafe_encode64
88.993k (± 5.5%) i/s - 887.796k in 10.009833s
Comparison:
NEW #urlsafe_encode64: 141104.0 i/s
OLD #urlsafe_encode64: 88993.5 i/s - 1.59x (± 0.00) slower

4 MB input
Warming up --------------------------------------
NEW #urlsafe_encode64
13.000 i/100ms
OLD #urlsafe_encode64
8.000 i/100ms
Calculating -------------------------------------
NEW #urlsafe_encode64
136.878 (± 2.9%) i/s - 1.378k in 10.074976s
OLD #urlsafe_encode64
87.759 (± 2.3%) i/s - 880.000 in 10.030723s
Comparison:
NEW #urlsafe_encode64: 136.9 i/s
OLD #urlsafe_encode64: 87.8 i/s - 1.56x (± 0.00) slower

[nobu]

str.chomp!("=") and str.chomp!("=") or str.chomp!("==") or str.chomp!("=") will work too.

[jcmfernandes]

str.chomp!("=") and str.chomp!("=") or str.chomp!("==") or str.chomp!("=") will work too.

Beautiful! 😍

EDIT:
I reran the benchmark and the performance is similar so I'm going with @nobu's idiomatic code.