Optimize label handling with predefined label keys in counter/gauge

[p0rtale]

This change adds support for predefined label keys in counter and gauge metric constructors, optimizing label processing.

Changes:

• Add optional label_keys table to counter() and gauge() initialization
• Replace dynamic label sorting with direct key generation when label_keys are provided, eliminating unnecessary processing overhead

Dynamic labels still supported when label_keys is omitted.

Closes TNTP-3453

[p0rtale]

variable labels: 83.25 s
fixed labels: 43.94 s

Bench script

local metrics = require('metrics')
local clock = require('clock')
local log = require('log')

local GAUGES_COUNT = 75
local KEYS_COUNT = 10000
local OPS_COUNT = 50000000

local label_keys = {
    "service", "version", "env", "region", "status", "deployment"
}

local label_values = {
    service = { "auth", "cart", "payment", "storage", "analytics", "notify", "gateway" },
    version = { "v1.0", "v1.1", "v1.2", "v2.0", "v2.1", "v2.2", "v3.0-beta" },
    env = { "prod", "stage", "dev", "test", "canary" },
    region = { "eu-west", "us-east", "us-west", "asia-south", "asia-east" },
    status = { "active", "inactive", "draining", "booting" },
    deployment = { "blue", "green", "red", "canary" }
}

local function generate_random_labels()
    local labels = {}
    for _, key in ipairs(label_keys) do
        local values = label_values[key]
        labels[key] = values[math.random(1, #values)]
    end
    return labels
end

local label_combinations = {}
for _ = 1, KEYS_COUNT do
    table.insert(label_combinations, generate_random_labels())
end

local var_gauges = {}
local fixed_gauges = {}
for i = 1, GAUGES_COUNT do
    var_gauges[i] = metrics.gauge('var_gauge_' .. i, 'Description for gauge ' .. i)
    fixed_gauges[i] = metrics.gauge('fixed_gauge_' .. i, 'Description for gauge ' .. i, {}, label_keys)
end

local function run_bench(gauges)
    local start_time = clock.monotonic()

    for _ = 1, OPS_COUNT do
        local gauge_idx = math.random(1, #gauges)
        local label_idx = math.random(1, #label_combinations)
        local value = math.random(1, 1000)

        gauges[gauge_idx]:set(value, label_combinations[label_idx])
    end

    return clock.monotonic() - start_time
end

local t1 = run_bench(var_gauges)
local t2 = run_bench(fixed_gauges)
local t3 = run_bench(var_gauges)
local t4 = run_bench(fixed_gauges)

log.info("variable labels: %.10f s", (t1 + t3) / 2)
log.info("fixed labels: %.10f s", (t2 + t4) / 2)

[DifferentialOrange]

What's the motivation? Is it the optimization or more assertive API for metrics users? If it's the optimization, I think the solution can be expanded to remove redundant table allocation (parts = {}) or even table.concat. As far as I remember, they also take a lot of work.

We shouldn't forget to

1. add a CHANGELOG entry,
2. update the API documentation (like here).