chore(halo): tune llama models.ini and drop 35B-A3B model

Serve only Qwen3.6-27B; remove the unused 35B-A3B preset.

Tuning:
- Move model-specific keys (spec-type, sampling temp/top-p/top-k/min-p)
  out of the [*] defaults into [Qwen3.6-27B] so they no longer leak onto
  other models; draft-mtp in particular only works on MTP-weighted models.
- Drop the duplicate parallel key from [*].
- Bump ubatch-size 256 -> 512 for faster iGPU prefill on Strix Halo.
- Add threads-batch = 16 to use all cores for prefill while keeping
  generation at threads = 8 under full GPU offload.

systems/x86_64-linux/halo/models.ini

@@ -6,30 +6,24 @@ parallel            = 1
 jinja               = true
 n-gpu-layers        = 99
 threads             = 8
-ubatch-size         = 256
+ubatch-size         = 512
 cache-type-k        = bf16
 cache-type-v        = bf16
-temp                = 0.6
-top-p               = 0.95
-top-k               = 20
-min-p               = 0.0
 mmap                = false
 no-context-shift    = true
 chat-template-kwargs = {"preserve_thinking": true}
 fit                 = on
-spec-type           = draft-mtp
-
-[Qwen3.6-35B-A3B]
-hf                  = unsloth/Qwen3.6-35B-A3B-MTP-GGUF:UD-Q8_K_XL
-spec-draft-n-max    = 2
-parallel            = 1
 c                   = 131072
-load-on-startup     = true
 
 [Qwen3.6-27B]
 hf                  = unsloth/Qwen3.6-27B-MTP-GGUF:Q6_K
+spec-type           = draft-mtp
 spec-draft-n-max    = 6
+threads-batch       = 16
+temp                = 0.6
+top-p               = 0.95
+top-k               = 20
+min-p               = 0.0
 parallel            = 2
 c                   = 524288
 load-on-startup     = true
-