perf(cas): LRU eviction instead of arbitrary HashMap iter order

[XciD]

Summary

When the reconstruction cache hits MAX_CACHE_ENTRIES and all remaining entries are full plans (range entries already purged in pass 1), the previous code picked the eviction victim via:

```rust
cache.keys().next().copied()
```

HashMap::keys() returns entries in hash-bucket order, which is unrelated to access patterns. A plan we just fetched and are actively reading from can be the first one returned, while an entry that's been cold for an hour survives.

This PR adds a last_accessed: Instant field to CacheEntry, bumped on every cache hit. The overflow path picks the entry with the oldest last_accessed. Hot plans stay warm even when they were inserted long ago.

Why not reuse inserted_at?

inserted_at is the time the CAS response arrived. Its only consumer is the TTL check — presigned URLs inside terms expire ~1h after the server issued them, not after we last touched the entry. Bumping inserted_at on hit would let stale URLs slip past the TTL.

Tests

New cache_evicts_lru_full_plan_on_overflow:

1. Fill cache 0..N with full plans.
2. Re-fetch entry 0 → bumps its last_accessed.
3. Insert one more entry → overflow.
4. Assert entry 0 still cached (no extra CAS call).
5. Assert entry 1 (true LRU) was evicted (extra CAS call on re-fetch).

All 328 lib + 338 nfs tests pass, clippy clean.

[github-actions]

POSIX Compliance (pjdfstest)

============================================================
  pjdfstest POSIX Compliance Results
------------------------------------------------------------
  Files: 130/130 passed    Tests: 832 total (0 subtests failed)
  Result: PASS
------------------------------------------------------------
  Category               Passed    Total   Status
  -------------------- -------- -------- --------
  chflags                     5        5       OK
  chmod                       8        8       OK
  chown                       6        6       OK
  ftruncate                  13       13       OK
  granular                    5        5       OK
  mkdir                       9        9       OK
  open                       19       19       OK
  posix_fallocate             1        1       OK
  rename                     10       10       OK
  rmdir                      11       11       OK
  symlink                    10       10       OK
  truncate                   13       13       OK
  unlink                     11       11       OK
  utimensat                   9        9       OK
============================================================

[github-actions]

Benchmark Results

============================================================
  Benchmark — 50MB
------------------------------------------------------------
  Metric                                 FUSE          NFS
  ------------------------------ ------------ ------------
  Sequential read                    268.7 MB/s     210.9 MB/s
  Sequential re-read                2263.3 MB/s    2444.3 MB/s
  Range read (1MB@25MB)                0.5 ms         0.2 ms
  Random reads (100x4KB avg)           0.0 ms         0.0 ms
  Sequential write (FUSE)           1308.7 MB/s
  Close latency (CAS+Hub)            0.101 s
  Write end-to-end                   358.3 MB/s
  Dedup write                       1663.9 MB/s
  Dedup close latency                0.113 s
  Dedup end-to-end                   348.8 MB/s
============================================================
============================================================
  Benchmark — 200MB
------------------------------------------------------------
  Metric                                 FUSE          NFS
  ------------------------------ ------------ ------------
  Sequential read                    757.5 MB/s     771.2 MB/s
  Sequential re-read                2260.6 MB/s    2458.7 MB/s
  Range read (1MB@25MB)                0.2 ms         0.2 ms
  Random reads (100x4KB avg)           0.0 ms         0.0 ms
  Sequential write (FUSE)           1593.3 MB/s
  Close latency (CAS+Hub)            0.183 s
  Write end-to-end                   648.4 MB/s
  Dedup write                       1620.3 MB/s
  Dedup close latency                0.115 s
  Dedup end-to-end                   837.4 MB/s
============================================================
============================================================
  Benchmark — 500MB
------------------------------------------------------------
  Metric                                 FUSE          NFS
  ------------------------------ ------------ ------------
  Sequential read                   1232.5 MB/s    1047.5 MB/s
  Sequential re-read                2260.5 MB/s    2508.2 MB/s
  Range read (1MB@25MB)                0.2 ms         0.2 ms
  Random reads (100x4KB avg)           0.0 ms         0.0 ms
  Sequential write (FUSE)           1512.3 MB/s
  Close latency (CAS+Hub)            0.330 s
  Write end-to-end                   757.2 MB/s
  Dedup write                       1540.1 MB/s
  Dedup close latency                0.154 s
  Dedup end-to-end                  1044.6 MB/s
============================================================
============================================================
  fio Benchmark Results
------------------------------------------------------------
  Job                        FUSE MB/s   NFS MB/s  FUSE IOPS   NFS IOPS
  ------------------------- ---------- ---------- ---------- ----------
  seq-read-100M                  473.9      531.9                      
  seq-reread-100M               2564.1      134.4                      
  rand-read-4k-100M                0.1        0.1         16         18
  seq-read-5x10M                 617.3      806.5                      
  rand-read-10x1M                  0.1        0.1         36         35
  Random Read Latency           FUSE avg      NFS avg
  ------------------------- ------------ ------------
  rand-read-4k-100M           63289.7 us   54491.4 us
  rand-read-10x1M             27946.4 us   28828.3 us
============================================================