Description

What original problem led to this PR?

PR #3704 added k.downcase calls in str_headers for Rack 3 compliance.
However, since we already call k.downcase in the case statement, we now
call it twice per header instead of once.

Are there related issues / prior discussions?

Follow-up optimization to PR #3704 (merged in v7.0.0) and issue #3250.

What alternatives have been tried?

Considered naming alternatives (k_lwr, lower_k) but settled on key
for readability.

Why do you make the choices you did? What are the tradeoffs?

• Use a local variable key to save the downcase result
• Zero behavior change
• Instead of calling downcase twice per header, we call it once. I did a GC.stat measurement and a micro-benchmark below.

Benchmark

Memory allocations

# 10 headers × 1000 iterations
GC.stat[:total_allocated_objects]

BEFORE: 20,004 allocations
AFTER:  10,004 allocations
GAIN:   10,000 allocations (50%)

Each String#downcase call creates a new String object. With ~10 headers
per response, this saves ~10 String allocations per request.

Micro-benchmark

                         user     system      total        real
BEFORE (2 downcase)  0.595148   0.003618   0.598766 (  0.674s)
AFTER  (1 downcase)  0.442885   0.002580   0.445465 (  0.490s)

In my runs, it was faster on the header loop (~27% in this instance), but I
wouldn't draw strong conclusions in terms of exact metrics.

Your checklist for this pull request

• I have reviewed the guidelines for contributing to this repository.
• I have added (or updated) appropriate tests if this PR fixes a bug or adds a feature.
• My pull request is 100 lines added/removed or less so that it can be easily reviewed.
• All new and existing tests passed, including Rubocop.