perf: standardize box_select benchmark

The output now contains items per second instead of time in nanoseconds.

Also, two new options are introduced:
* The `--output` option allows you to specify the output file.
* The `--output_format` option means the format for the printed output.
  The default is "console". It just prints the number of iterations
  proceeded per second to the stdout. The "json" format contains all the
  information about the benchmark in a format similar to Google
  Benchmark's.

Usually, these options should be used together to dump machine-readable
results for the benchmarks.

NO_DOC=perf test
NO_CHANGELOG=perf test
NO_TEST=perf test

perf/lua/box_select.lua

 --
 -- The test measures run time of various space select operations.
 --
--- Output format:
--- <test-case> <run-time-nanoseconds>
---
--- Options:
--- --pattern <string>  run only tests matching the pattern; it's possible
---                     to specify more than one pattern separated by '|',
---                     for example, 'get|select'
--- --read_view         use a read view
---
+-- Output format (console):
+-- <test-case> <iterations-per-second>
 
 local clock = require('clock')
 local fiber = require('fiber')
+local benchmark = require('benchmark')
+
+local USAGE = [[
+   pattern <string>           - run only tests matching the pattern; it's
+                                possible to specify more than one pattern
+                                separated by '|', for example, 'get|select'
+   read_view <boolean, false> - use a read view
 
-local params = require('internal.argparse').parse(arg, {
+ Being run without options, this benchmark measures the performance of
+ various space select (get, select, pairs) operations.
+]]
+
+local params = benchmark.argparse(arg, {
     {'pattern', 'string'},
     {'read_view', 'boolean'},
-})
+}, USAGE)
+
 if params.pattern then
     params.pattern = string.split(params.pattern, '|')
 end
 
+local bench = benchmark.new(params)
+
 box.cfg({log_level = 'error'})
 box.once('perf_select_init', function()
     local s = box.schema.space.create('perf_select_space')
@@ -222,11 +229,20 @@ local TESTS = {
     },
 }
 
+local function single_run(func)
+    local real_time_start = clock.time()
+    local cpu_time_start = clock.proc()
+    func()
+    local delta_real = clock.time() - real_time_start
+    local delta_cpu = clock.proc() - cpu_time_start
+    return delta_real, delta_cpu
+end
+
 --
 -- Runs the given test case function in a loop.
 -- Returns the average time it takes to run the function once.
 --
-local function bench(test)
+local function run_bench(test)
     local warmup_runs = 1e2 / 4
     local test_runs = 1e2
     local iters_per_run = 1e4
@@ -236,19 +252,26 @@ local function bench(test)
             func()
         end
     end
-    local test_time = 0
+    local real_time = 0
+    local cpu_time = 0
     for i = 1, warmup_runs + test_runs do
         for _ = 1, 5 do
             collectgarbage('collect')
             jit.flush()
             fiber.yield()
         end
-        local t = clock.bench(run)[1]
+        local delta_real, delta_cpu = single_run(run)
         if i > warmup_runs then
-            test_time = test_time + t
+            real_time = real_time + delta_real
+            cpu_time = cpu_time + delta_cpu
         end
     end
-    return test_time / test_runs / iters_per_run
+    local iterations = test_runs * iters_per_run
+    bench:add_result(test.name, {
+        real_time = real_time,
+        cpu_time = cpu_time,
+        items = iterations
+    })
 end
 
 for _, test in ipairs(TESTS) do
@@ -263,9 +286,10 @@ for _, test in ipairs(TESTS) do
         end
     end
     if not skip then
-        local t = bench(test)
-        print(string.format('%s %d', test.name, t * 1e9))
+        run_bench(test)
     end
 end
 
+bench:dump_results()
+
 os.exit(0)