Cache symbol completion result

[tompng]

Make symbol completion faster

• Cache Symbol.all_symbols.sort and reuse it.
• Restrict the number of symbol candidates to reduce completion/rendering cost in Reline.
• Use bsearch to perform O(logN) search from cached sorted symbol list.

irb -f
irb(main):001> s='a';symbols = 100000.times.map{s=s.succ; [s.to_sym, ('a'+s).to_sym]};
irb(main):002> :aaabq
               :aaabk 
               :aaabl 
               :aaabm 
               :aaabn 
               :aaabo 
               :aaabp 
               :aaabq▄
               :azyd █
               :azye ▀
               :azyf  
               :azyg  
               :azyh  
               :azyi  
               :azyj  
               :azyk  

Completion candidates of symbol contains first 50 + last 50 symbols matched to the completion target.

[tompng]

I think there is room for discussion whether to keep symbol completion or to drop it.

If we're going to keep it, we need to do this bsearch because symbol filtering (main bottleneck in IRB side) is few times slower than Symbol.all_symbols.

[tompng]

In the following situation, Symbol.all_symbol.sort takes about a second. I think it is not acceptable.

Symbol.all_symbols.size is 100000 with average symbol length=15 (About the number in Rails app) Running in docker (host and image platform mismatch)