Fix configured PHP benchmark workload reuse

[chubes4]

Summary

• load each configured PHP workload file once per benchmark command
• reuse its returned callable across warmup and measured iterations
• cover named helper declarations and per-iteration callable execution

Verification

• npm exec -- tsx tests/bench-command-step-behavior.test.ts
• npm run build
• git diff --check

Fixes #2446.

AI assistance

GPT-5.6 Sol via OpenCode diagnosed the configured-workload lifecycle mismatch from retained homeboy-lab evidence, implemented the focused repair and regression coverage under Chris Huber’s direction, and ran the listed verification.