Adopt five open benchmark PRs

[JuanVqz]

Summary

• Lands five open benchmark PRs, one commit each, with every contributor credited as co-author, so each contribution can be read (and rejected) on its own.
• Each one is brought up to today's rules: winner named by rank methods and listed first, runs on every Ruby in CI, same work in every report.
• README entries keep the contributor's own text and sample output, and add a run on Ruby 4.0.7 below it.

Changes

Step 1: Hash#values_at vs Array#map { Hash#[] } (#174 by @kewlar)

• The file as written, without the spaces padding one report label.
• Local runs: values_at wins on Ruby 2.1, 3.4, 4.0, head and JRuby head (1.52x to 2.10x). On TruffleRuby head map was 1.06x ahead in one run, so no comment until CI confirms it.

Step 2: method vs forwarded method vs delegated method (#182 by @yogeshjain999)

• The delegated report called pop(1) while the other two push, so it measured different work. It now calls << 1, which SimpleDelegator passes to the array, and all three return [1].
• slowest becomes slower, so the rank names do not skip a step.
• The PR's Ruby 2.2 sample is kept, with a note that it ran the pop version.

Step 3: Array#sum vs Array#inject(:+) (#140 by @jrmhaig)

• Wrapped in if RUBY_VERSION >= '2.4.0', as suggested on the PR, so it skips Rubies without Array#sum instead of crashing.
• Drops require "rubygems" and the .travis.yml change (main no longer has Travis). Labels Array#sum and Array#inject(:+).
• A comment that the two only agree for Integers: for Floats, sum compensates for rounding errors.
• The PR had no README entry, so this adds one next to the inject entry.

Step 4: array subset check (#237 by @etagwerker, from #125 by @gabteles)

• Rank-named methods instead of minus_empty, all_include and so on. (a1 - a2).empty? is the winner, as the PR found on 3.3, 3.4 and 4.0.
• A comment on where the winner changes: all? + include? wins on JRuby and TruffleRuby, and ties on Ruby 2.1.

Step 5: non-nil hash values, values.compact vs select (#236 by @etagwerker, from #124 by @sFrenkie)

• No more Random.rand in the data, so the number of nils is the same on every run: every other value is nil.
• The two select reports keep values with !v.nil? instead of by truthiness, so like compact they drop only nil and keep false (from the PR's Copilot review). With { a: false, b: nil, c: 1 } all three now return [false, 1]; the old select { |v| v } returned [1]. The README text's two snippets change the same way.
• Rank-named methods, and the sanity check compares results without sorting (also from the Copilot review).
• On ruby_4.0.7 values.compact is now 21.05x faster (13.78x before !v.nil?, which is a little more work than v). The PR's ruby_4.0.0 sample, which ran the truthiness version, is kept.

Test plan

• The lint passes all 73 files, and RuboCop's quote check passes on the five new files.
• Steps 1, 2, 4 and 5 ran on Ruby 2.1, 3.4, 4.0, head, JRuby head and TruffleRuby head (Docker, one run each). Step 5's six-Ruby run was before its data and !v.nil? changes; after them, it ran on 4.0.7.
• Step 3 skips cleanly on Ruby 2.1, 2.3 and JRuby 9.1, and runs on 2.4, 4.0 and TruffleRuby 22.
• Each file's sanity check (or a direct check) confirms its reports return the same result.
• CI runs only the five new files, on all 26 jobs.

After this merges, each contributor PR gets a comment saying where it landed, and is closed.