(perf) introduce perf test harness and refactor (#54)

Made a lot of backwards-compatible changes and simplifications to the code that I had wanted to do for a lot of time, while also keeping an eye in performance. The new changes improved overall performance on my limited, dev-machine-only, luajit-only tests.

Before:

```
$ luajit perf.lua
nil,string,empty,sequence,record,hybrid,recursive,meta,process,complex,big
0.054246,0.140177,0.149759,0.323062,0.441312,0.607064,0.296031,0.458631,0.105668,1.717528,2.047272
```

After:

```
$ luajit perf.lua
nil,string,empty,sequence,record,hybrid,recursive,meta,process,complex,big
0.036804,0.188984,0.099349,0.247425,0.342079,0.452339,0.203801,0.288931,0.169462,1.010763,1.616462
```

perf.lua

@@ -0,0 +1,112 @@
+local inspect = require 'inspect'
+
+local skip_headers = ...
+
+local N=100000
+
+local results = {}
+
+local time = function(name, n, f)
+  local clock = os.clock
+
+  collectgarbage()
+  collectgarbage()
+  collectgarbage()
+
+  local startTime = clock()
+
+  for i=0,n do f() end
+
+  local duration = clock() - startTime
+
+  results[#results + 1] = { name, duration }
+end
+
+-------------------
+
+time('nil', N, function()
+  inspect(nil)
+end)
+
+time('string', N, function()
+  inspect("hello")
+end)
+
+local e={}
+time('empty', N, function()
+  inspect(e)
+end)
+
+local seq={1,2,3,4}
+time('sequence', N, function()
+  inspect(seq)
+end)
+
+local record={a=1, b=2, c=3}
+time('record', N, function()
+  inspect(record)
+end)
+
+local hybrid={1, 2, 3, a=1, b=2, c=3}
+time('hybrid', N, function()
+  inspect(hybrid)
+end)
+
+local recursive = {}
+recursive.x = recursive
+time('recursive', N, function()
+  inspect(recursive)
+end)
+
+local with_meta=setmetatable({},
+  { __tostring = function() return "s" end })
+time('meta', N, function()
+  inspect(with_meta)
+end)
+
+local process_options = {
+  process = function(i,p) return "p" end
+}
+time('process', N, function()
+  inspect(seq, process_options)
+end)
+
+local complex = {
+  a = 1,
+  true,
+  print,
+  [print] = print,
+  [{}] = { {}, 3, b = {x = 42} }
+}
+complex.x = complex
+setmetatable(complex, complex)
+time('complex', N, function()
+  inspect(complex)
+end)
+
+local big = {}
+for i = 1,1000 do
+  big[i] = i
+end
+for i = 1,1000 do
+  big["a" .. i] = 1
+end
+time('big', N/100, function()
+  inspect(big)
+end)
+
+------
+
+if not skip_headers then
+  for i,r in ipairs(results) do
+    if i > 1 then io.write(",") end
+    io.write(r[1])
+  end
+  io.write("\n")
+end
+
+for i,r in ipairs(results) do
+  if i > 1 then io.write(",") end
+  io.write(r[2])
+end
+io.write("\n")