fix(descriptor): use _normalize_header_name in header fast-path

[levleontiev]

Summary

The two-step fast-path for header lookup in descriptor.lua was applying only one transformation at a time, while OpenResty applies both simultaneously (lowercase + hyphen→underscore):

-- BEFORE (broken for uppercase+hyphenated names):
local name_underscore = string_gsub(name, "-", "_")  -- e.g. X-E2E-Key → X_E2E_Key (wrong: case preserved)
local name_lower      = string_lower(name)            -- e.g. X-E2E-Key → x-e2e-key (wrong: hyphens preserved)
value = headers[name_underscore] or headers[name_lower]
-- Both miss when headers dict is OpenResty-normalized: {"x_e2e_key": "val"}
-- Falls through to O(n) pairs() scan every time

Fix

Replace with a single call to the already-defined _normalize_header_name helper that applies both transformations:

-- AFTER (correct):
value = headers[_normalize_header_name(name)]
-- _normalize_header_name("X-E2E-Key") = "x_e2e_key" → O(1) hit

Impact

• Any limit key using an uppercase+hyphenated header name (e.g. header:X-E2E-Key, header:X-API-Key) with an OpenResty-normalized header table was doing an O(n) pairs() scan on every request instead of an O(1) lookup.
• Correctness was maintained (fallback scan found the value), but performance degraded linearly with response header count.

Why tests didn't catch this

AC-7b uses a lowercase limit key (header:x-e2e-key), so string_gsub("x-e2e-key", "-", "_") = "x_e2e_key" — the fast-path happened to work. AC-7c (new) covers the uppercase case.

🤖 Generated with Claude Code