Dual-SSD streaming: COLI_MODEL_MIRROR reads experts from two model copies

A second (read-only) copy of the model on another drive is registered as a
per-shard read replica; expert loads are split between the drives by a
deterministic (layer,eid) hash weighted by COLI_DISK_WEIGHTS=<primary>,<mirror>
or, by default, by a startup bandwidth probe using the engine's own access
pattern (parallel ~19 MB reads, O_DIRECT). Cold decode is disk-bound
(~11 GB/token), so two NVMe queues add up.

- st.h: mirror accepted per file only if size + safetensors header are
  byte-identical to the primary (identical data_offsets by construction, so
  every pread is valid on either copy); partial mirrors work (smaller second
  SSD holding only some shards); the mirror is never written — .coli_usage /
  .coli_kv stay on the primary.
- glm.c: routing covers the coalesced slab pread, O_DIRECT, mmap/Metal and
  scale reads, plus the OMP-parallel pin/autopin warmup (streams from both
  drives). Deterministic routing keeps readahead/PILOT WILLNEED on the same
  drive as the demand read and avoids caching an expert twice. A mirror read
  error falls back to the primary (one warning, no crash). Per-drive bytes
  are reported in a MIRROR: stats line.
- tests: test_st_mirror covers validation, read equality on both replicas,
  and the rejection paths (divergent header, size mismatch, missing file).

Measured on 2x NVMe (GLM-5.2 int4, greedy, DRAFT=0, DIRECT=1, cold-ish
cache): 0.42 -> 0.57 tok/s (+36%), expert-disk service 15.2s -> 10.3s,
byte-identical output; probe chose a 48/52 split.

c/coli

@@ -15,6 +15,9 @@ Run GLM-5.2 (744B) locally on CPU with roughly 15-26 GB of RAM.
 
 Configuration through environment variables or flags (also valid after the subcommand):
   COLI_MODEL=<dir>   model directory (default /home/vincenzo/glm52_i4)
+  COLI_MODEL_MIRROR=<dir>  second copy of the model on another drive: expert reads
+                     are split across both SSDs (COLI_DISK_WEIGHTS=9,3 sets the
+                     primary,mirror bandwidth ratio; default: measured at startup)
   --ram N            RAM budget in GB (automatically sizes the expert cache)
   --repin N          adapt RAM/VRAM experts every N tokens
   --topp P           adaptive expert top-p             --topk N   fixed top-k