Avoid repeated conversions when shifting Calendar.ISO times

[preciz]

Assisted-by: Codex:GPT-6

~1.5x faster, uses less memory.

Replace duration options and reduction with tuple dispatch. Fold whole
seconds into the time tuple and apply microseconds in a single shift,
preserving precision and a fast path for zero shifts.

Bench:

# Run with: bin/elixir bench.exs
Mix.install([:benchee])

path = "lib/elixir/lib/calendar/iso.ex"

for {ref, module} <- [{"main", Old}, {"HEAD", New}] do
  {source, 0} = System.cmd("git", ["show", "#{ref}:#{path}"], cd: __DIR__)

  source
  |> String.replace("defmodule Calendar.ISO do", "defmodule #{inspect(module)} do")
  |> Code.compile_string(path)
end

inputs = %{
  "zero duration" => [],
  "seconds only" => [second: 86_401],
  "microseconds only" => [microsecond: {500_001, 6}],
  "seconds and microseconds" => [second: 86_401, microsecond: {500_001, 6}]
}

Benchee.run(
  %{
    "main" => &Old.shift_time(23, 59, 59, {120_000, 3}, &1),
    "HEAD" => &New.shift_time(23, 59, 59, {120_000, 3}, &1)
  },
  inputs: Map.new(inputs, fn {name, units} -> {name, Duration.new!(units)} end),
  pre_check: :all_same,
  warmup: 1,
  time: 3,
  memory_time: 1
)

Results:

Operating System: Linux
CPU Information: AMD Ryzen 7 8845HS w
Number of Available Cores: 16
Available memory: 54.72 GB
Elixir 1.21.0-dev
Erlang 29.0.5
JIT enabled: true

Benchmark suite executing with the following configuration:
warmup: 1 s
time: 3 s
memory time: 1 s
reduction time: 0 ns
parallel: 1
inputs: microseconds only, seconds and microseconds, seconds only, zero duration
Estimated total run time: 40 s
Excluding outliers: false

##### With input microseconds only #####
Name           ips        average  deviation         median         99th %
HEAD        9.91 M      100.89 ns  ±5244.69%          90 ns         131 ns
main        7.92 M      126.30 ns  ±3650.24%         101 ns         201 ns

Comparison: 
HEAD        9.91 M
main        7.92 M - 1.25x slower +25.40 ns

Memory usage statistics:

Name    Memory usage
HEAD           288 B
main           368 B - 1.28x memory usage +80 B

**All measurements for memory usage were the same**

##### With input seconds and microseconds #####
Name           ips        average  deviation         median         99th %
HEAD        9.33 M      107.20 ns  ±5579.27%          90 ns         140 ns
main        4.53 M      220.86 ns  ±3458.13%         190 ns         310 ns

Comparison: 
HEAD        9.33 M
main        4.53 M - 2.06x slower +113.67 ns

Memory usage statistics:

Name    Memory usage
HEAD           288 B
main           640 B - 2.22x memory usage +352 B

**All measurements for memory usage were the same**

##### With input seconds only #####
Name           ips        average  deviation         median         99th %
HEAD        8.82 M      113.42 ns  ±5136.35%          90 ns         180 ns
main        7.13 M      140.33 ns  ±4783.83%         110 ns         220 ns

Comparison: 
HEAD        8.82 M
main        7.13 M - 1.24x slower +26.91 ns

Memory usage statistics:

Name    Memory usage
HEAD           312 B
main           392 B - 1.26x memory usage +80 B

**All measurements for memory usage were the same**

##### With input zero duration #####
Name           ips        average  deviation         median         99th %
HEAD       29.05 M       34.43 ns  ±4411.93%          30 ns          41 ns
main       18.05 M       55.41 ns  ±1479.18%          50 ns          81 ns

Comparison: 
HEAD       29.05 M
main       18.05 M - 1.61x slower +20.98 ns

Memory usage statistics:

Name    Memory usage
HEAD            40 B
main           120 B - 3.00x memory usage +80 B

**All measurements for memory usage were the same**

[josevalim]

💚 💙 💜 💛 ❤️