Iterate binding table using standard 'for' loop

[dpino]

After the use of recycle the only interpreted code reported by LuaJIT's profiler is on lwaftr.lua:

• -jp=vF (per VM states and files).

35%  Interpreted
  -- 48%  lwaftr.lua:method
  -- 24%  lwaftr.lua:binding_lookup_ipv4_from_pkt
  -- 22%  lwaftr.lua:_encapsulate_ipv4

• -jp=Fl (per file and line)

32%  lwaftr.lua:binding_lookup_ipv4_from_pkt
  -- 36%  lwaftr.lua:63
  -- 33%  lwaftr.lua:67
  -- 15%  lwaftr.lua:68
  --  8%  lwaftr.lua:65
  --  4%  lwaftr.lua:43
31%  lwaftr.lua:method
  -- 49%  lwaftr.lua:362
  -- 22%  lwaftr.lua:366
  -- 11%  lwaftr.lua:367
  --  6%  lwaftr.lua:358
  --  4%  lwaftr.lua:379
  --  3%  lwaftr.lua:363
  --  3%  lwaftr.lua:374
19%  lwaftr.lua:_encapsulate_ipv4
  -- 28%  lwaftr.lua:262
  -- 26%  lwaftr.lua:263
  --  7%  lwaftr.lua:278
  --  6%  lwaftr.lua:242
  --  5%  lwaftr.lua:190
  --  5%  lwaftr.lua:276
  --  4%  lwaftr.lua:265
  --  3%  lwaftr.lua:290

Entry -- 4% lwaftr.lua:43 corresponds to a loop in binding_lookup_ipv4_from_pkt. Iterating the binding table with a standard for loop instead of ipairs makes LuaJIT happy and after the patch all interpreted code is gone.

99%  Compiled
  -- 12%  class.lua:free
  -- 12%  packet.lua:clone
  -- 10%  packet.lua:prepend
  --  7%  class.lua:new
  --  6%  packet.lua:shiftleft
  --  5%  datagram.lua:push
  --  5%  link.lua:receive
  --  4%  basic_apps.lua:method
  --  3%  lwaftr.lua:binding_lookup_ipv4_from_pkt

Performance improves significantly:

Processed 10.9 million packets in 5.00 seconds (7016299500 bytes; 11.23 Gbps)
Made 42,725 breaths: 255.00 packets per breath; 117.03 us per breath
Rate(Mpps): 2.179

I run the end-to-end.sh tests. All tests passed.

[kbara]

Merging; that's a good performance number to see, and some nice profiling work. That said, in practice we'll be using a hash table here; looping over the whole table is something that only makes sense for toy tables.