Add directory-level locking and introduce layered index cache

Introduces directory-level locking to prevent concurrent index operations from corrupting shared directories, along with explicit APIs for acquiring, probing, and releasing locks. Restructures the index cache crate to use a layered store architecture that eagerly initializes metadata and provides fast hierarchical lookups. Updates dependent modules, test suites, and CLI commands to align with the refactored API surface, and adds an end-to-end smoke test for validation.

src/obikmer2/src/cmd/index/mod.rs

@@ -168,16 +168,6 @@ pub fn run(args: IndexArgs) {
     let output = args.output.clone();
     let mut rep = Reporter::new();
 
-    // Locked for the whole build (including a possible --force removal +
-    // recreation below): a second `index` run resuming/overwriting the same
-    // output directory concurrently would otherwise corrupt it. Unlinking
-    // the lock file via --force's remove_dir_all is safe — the held file
-    // descriptor keeps the lock regardless of the directory entry.
-    let _lock = obisys::DirLock::acquire(&output).unwrap_or_else(|e| {
-        eprintln!("error locking output directory {}: {e}", output.display());
-        std::process::exit(1);
-    });
-
     // ── Resolve evidence kind ────────────────────────────────────────────────
     let (evidence, effective_kmer_size) = if args.approx {
         let (z, b, fp) = resolve_approx_params(args.findere_z, args.evidence_bits, args.fp);