fix(build): reclaim expired prometheus shared dict entries

[wklken]

Description

Backport the physical shared-dict reclamation fix from api7/nginx-lua-prometheus#18.

The existing cleanup removes only worker-local key-index references. Expired metric values and index slots can continue occupying shared-dict slab pages under label churn, causing prometheus-metrics usage to grow without recovering. The build patch calls flush_expired() after the existing expiry scan.

Related:

• bug: Prometheus shared dict free_space drops periodically when metric expire is enabled apache/apisix#13658
• fix(keys): reclaim expired shared-dict entries in remove_expired_keys() api7/nginx-lua-prometheus#18

Verification

• Patch dry-run with --fuzz=0 against both apache/apisix:3.16.0-redhat and apache/apisix:3.17.0-redhat
• Built bk-apigateway-apisix:prometheus-flush-master-codex from this branch
• Verified the built APISIX 3.17.0 image contains exactly one self.dict:flush_expired() call in prometheus_keys.lua
• LuaJIT bytecode compilation of the patched file succeeded
• RUN_WITH_IT= make lint: 0 warnings / 0 errors
• RUN_WITH_IT= make test: Busted 752 successes; test-nginx 681 tests, all successful

Checklist

• 填写 PR 描述及相关 issue (write PR description and related issue)
• 代码风格检查通过 (code style check passed)
• PR 中包含单元测试 (include unit test) — dependency patch; upstream PR contains regression tests
• 单元测试通过 (unit test passed)
• 本地开发联调环境验证通过 (local development environment verification passed)