fix: fxcache key uses mtime which changes on PVC remount — use size only

The cache key hashed filename + size + mtime. PVC remounts on different
nodes change file timestamps without changing data, producing a different
key every pod boot. Result: fxcache MISS on every H100 run, forcing
10+ min DBN feature extraction from 148GB raw MBP-10 data.

Fix: hash filename + size only. Deterministic across remounts.
Note: existing cache must be rebuilt once (key format changed).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/ml/src/feature_cache.rs

@@ -7,7 +7,6 @@
 use anyhow::{Context, Result};
 use sha2::{Digest, Sha256};
 use std::path::{Path, PathBuf};
-use std::time::SystemTime;
 
 /// Calculate a cache key for a DBN data directory.
 ///
@@ -52,7 +51,9 @@ pub fn calculate_dbn_cache_key_full(
     }
 
     for path in &files {
-        // Hash filename (not full path) + size + mtime
+        // Hash filename (not full path) + size.
+        // NOT mtime: PVC remounts and node reboots change file timestamps
+        // without changing data, causing cache misses on every new pod.
         if let Some(name) = path.file_name() {
             hasher.update(name.to_string_lossy().as_bytes());
         }
@@ -60,13 +61,6 @@ pub fn calculate_dbn_cache_key_full(
             .metadata()
             .with_context(|| format!("Failed to stat {:?}", path))?;
         hasher.update(&meta.len().to_le_bytes());
-        let mtime = meta
-            .modified()
-            .context("mtime unavailable")?
-            .duration_since(SystemTime::UNIX_EPOCH)
-            .context("mtime before epoch")?
-            .as_secs();
-        hasher.update(&mtime.to_le_bytes());
     }
 
     Ok(format!("{:x}", hasher.finalize()))