Improve control performance

[Benjamin Klum]

Now quick and long fader moves sometimes lead to audio crackling.

[Benjamin Klum]

re #115 Make dispatch to main thread faster

→ <<cset 85aa4b000905>>

[Benjamin Klum]

re #115 Don't allocate vector when MIDI message arrives

Instead use the fast queue mechanism to pass each target value change operation to the main thread separately.

→ <<cset 52bf4223fcf6>>

[Benjamin Klum]

re #115 Don't allocate source object on heap when MIDI message arrives

Instead don't create a source object at all but just ask each mapping if the passed message fits to the mapping's source.

→ <<cset 8806864b9b0c>>

[Benjamin Klum]

re #115 Don't allocate vector when forwarding (N)RPN/14-bit messages

→ <<cset e8cfa1aca19d>>

[Benjamin Klum]

re #115 Don't allocate lambda when MIDI message arrives

That should get rid of the last dynamic allocation in ReaLearn's real-time thread code.

→ <<cset 97bb5e712600>>