Use a binary accumulator in QuotedPrintable encoder to reduce memory usage

[michallepicki]

Hi! I noticed huge memory usage when encoding big emails (~15MB, that big mainly because of someone's huge base64-encoded inline image in their mail signature) after investigation, I found that currently the code builds a list of small binaries, reverses and joins it. As per the Erlang Efficiency Guide, appending to binaries is well optimized and should perform better.

I also confirmed the functions can be tail-call optimized the way they're written currently, with returning if expressions (at least on newest OTP 25).

Similarly to #86 , here is the benchmark used and results:

len = 1024 * 1024 * 15
bin =
  (len * 2)
  |> :crypto.strong_rand_bytes()
  |> Base.encode64()
  |> String.slice(0, len)

Benchee.run(%{
  "OldQuotedPrintable.encode/1" => fn ->
    Mail.Encoders.OldQuotedPrintable.encode(bin)
  end,
  "BinaryAccQuotedPrintable.encode/1" => fn ->
    Mail.Encoders.QuotedPrintable.encode(bin)
  end,
}, time: 10, memory_time: 2)

Results:

# mix run bench.exs
Operating System: Linux
CPU Information: AMD Ryzen 9 5900HX with Radeon Graphics
Number of Available Cores: 16
Available memory: 13.58 GB
Elixir 1.14.0-dev
Erlang 25.0

Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 2 s
reduction time: 0 ns
parallel: 1
inputs: none specified
Estimated total run time: 28 s

Benchmarking BinaryAccQuotedPrintable.encode/1 ...
Benchmarking OldQuotedPrintable.encode/1 ...

Name                                        ips        average  deviation         median         99th %
BinaryAccQuotedPrintable.encode/1          2.73         0.37 s     ±2.48%         0.37 s         0.38 s
OldQuotedPrintable.encode/1                0.23         4.29 s     ±0.93%         4.27 s         4.34 s

Comparison: 
BinaryAccQuotedPrintable.encode/1          2.73
OldQuotedPrintable.encode/1                0.23 - 11.73x slower +3.93 s

Memory usage statistics:

Name                                 Memory usage
BinaryAccQuotedPrintable.encode/1         0.59 GB
OldQuotedPrintable.encode/1               1.06 GB - 1.81x memory usage +0.48 GB

**All measurements for memory usage were the same**

[bcardarella]

Nice!