[P0-3/P0-4 fix] Route hydra_config bootstrap through CONFIGURE(0x40) task queue

[hydra-z]

Fix for #49 P0-3 and P0-4

Problem

P0-3: meta is never populated after inline hydra_config first load — update_meta() only called at startup, not after apply_t3_rebuild() in the inline path.

P0-4: apply_t3_rebuild() is called from HTTP worker thread (line 6669) without mutex, racing the main task-queue thread that accesses the same ctx_tgt/model_tgt/slots state.

Fix

Remove the inline HTTP bootstrap from post_chat_completions entirely. Instead:

1. Remove the && ctx_tgt guard in process_single_task() (server-context.cpp:3259,3265) — now unnecessary since fork: hydra_config inline T3 + hydra_metrics response (hydra_vortex #411) #48 made T3 statics ctx-independent
2. Enqueue a CONFIGURE(0x40) task instead of adding a second synchronous HTTP-thread path
3. This fixes P0-3 (meta gets populated via existing CONFIGURE path), P0-4 (no more HTTP-thread race), and P1-5 (no more shared hydra_metrics_result)

Files to modify

• tools/server/server-context.cpp — Remove inline hydra_config block from post_chat_completions (lines 6633-6689)
• tools/server/server-context.cpp — Remove && ctx_tgt guard in CONFIGURE handler (lines 3259,3265)

Verification

• Build: cmake --build build_sm120 --target llama-engine
• Test: Start engine without --model, send CONFIGURE with model_path
• Check: meta populated, no null-deref, no race under concurrent load

E2E confirmation

Source analysis confirmed: inline path bypasses task queue, races main thread, never populates meta. CONFIGURE(0x40) path is the documented, queue-mediated mechanism that already works on both binaries.

Cross-repo

• Hydra parent: [review] PR #48 (hydra_config inline T3) — architecture gaps in llama-engine COMBINE mode bring-up hydra_vortex#415
• Companion fix: P0-2 (dangling pointer)
• Related: [P2] Cleanup: hydra_config CONFIGURE path duplicates existing mechanisms, dead code, wasted copies #50 P2-7 (altitude recommendation)