[personality] support harness-only personality none

[rhan-oai]

Why

The no-personality harness mode previously only worked through the legacy {{ personality }} template. Model prompts that bake personality guidance into a # Personality section need source-aware post-processing instead.

Only catalog-provided instructions are eligible for stripping. Explicit instructions remain authoritative, and the integration test verifies that a custom prompt containing # Personality reaches the outbound request byte-for-byte unchanged.

What changed

• post-process catalog-provided model instructions only for explicit Some(Personality::None)
• remove the first exact # Personality section through, but not including, the next H1
• preserve explicit base-instruction overrides, resumed prompts, additive developer instructions, and existing Friendly/Pragmatic behavior
• treat missing personality metadata as a valid fallback for Personality::None without emitting the named-personality warning
• add table-driven models-manager coverage and focused outbound core integration coverage

Testing

• just test -p codex-models-manager
• just test -p codex-protocol
• just test -p codex-core config_personality_none_
• locally validated for 5.6 sol