ops: Signal disk-full and eviction capacity pressure

When the disk (or memory) tier is at cap or the volume returns ENOSPC,
ops currently look like random misses with no clear "we are dropping
data." Count those events as capacity_pressure_events on /metrics and
log tier plus reason so operators can tell capacity pressure from a
cold cache, without changing the existing evictions counter.

Link: #36

README.md

@@ -85,6 +85,7 @@ curl -s -i http://localhost/lancache-heartbeat
 | `memory_cache_hits` / `disk_cache_hits` | Which tier served the hits |
 | `total_requests` / `errors` | Volume and failures |
 | `disk_tier_ready` | `0` while disk slow-tier attach pending; `1` when attached, or when no disk configured (N/A — not waiting) |
+| `capacity_pressure_events` | Soft eviction under the memory or disk cap, and/or disk Create/Write/Mkdir hitting ENOSPC (volume full). Distinct from cold-cache misses and from the existing `evictions` counter. Logs `tier` (memory or disk) and `reason` (eviction or enospc). |
 
 A first pass through new content is mostly misses (`hit_rate` near 0). Repeat the same content and `cache_hits` / `hit_rate` should rise.
 
@@ -296,6 +297,7 @@ See `config.Validate()` and `steamcache.New` error paths. This ensures the LAN a
 - Startup logs: Info "Disk slow tier attach pending..." then later "Disk slow tier attached (...)" for disk-only and mixed modes.
 - `/metrics` exposes `disk_tier_ready` 0/1 and stays responsive during attach (GetMetrics does not block on Size while pending).
 - `/lancache-heartbeat` header `X-SteamCache-Disk-Tier` mirrors that state.
+- `/metrics` `capacity_pressure_events` counts times the cache dropped data under capacity pressure (soft eviction at the memory or disk cap, or disk Create/Write/Mkdir returning ENOSPC). Logs include `tier=memory|disk` and `reason=eviction|enospc` so operators can grep and tell this apart from a cold cache. The existing `evictions` counter is unchanged.
 
 #### Garbage Collection Algorithms