Use Process.on_interrupt instead of Signal::INT.trap

Instead of Signal::INT.trap, Toro now uses the platform-independent Process.on_interrupt.

A new way of configuring the server

The server no longer listens by default, and the call to
run does not require a port number to be passed. This
allows for more flexibility when configuring the server.

If App is an instance of Toro, then a call to App.run will
yield an instance of HTTP://Server that you can configure.

For example, you can start the server on port 80:

App.run do |server|
  server.listen "0.0.0.0", 80
end

The following example shows how to configure SSL certificates:

App.run do |server|
  ssl = OpenSSL::SSL::Context::Server.new
  ssl.private_key = "path/to/private_key"
  ssl.certificate_chain = "path/to/certificate_chain"
  server.tls = ssl
  server.listen "0.0.0.0", 443
end

Configurable Server

If App is an instance of Toro, then you can start the server
by calling App.run. You can pass any options you would use with
the HTTP::Server constructor from the standard library.

For example, you can start the server on port 80:

App.run(80)

Or you can further configure server by using a block. The following
example shows how to configure SSL certificates:

App.run(443) do |server|
  ssl = OpenSSL::SSL::Context::Server.new
  ssl.private_key = "path/to/private_key"
  ssl.certificate_chain = "path/to/certificate_chain"
  server.tls = ssl
end

Refer to Crystal's documentation for more options.

JSON helper

Toro gains a json helper. If you require "json" from the standard library, you can use the json helper with any generic Crystal object. Thanks @RayDF for this feature!

Ridiculous speedup

I felt Toro was fast, but then I got a pull request from @RayDF where he got rid of the workflow around raising the Halt exception, something Toro was using for stopping the execution once a match was found. Instead, he added a flag (the @halt instance variable) to keep track if a match had occurred. The result was an increase in performance of around 40%. The final version doesn't need the flag, as we were able to replace part of the API with a set of macros and the behavior is more in line with that of the first release. Huge thanks to @RayDF for his awesome work :-)