[roll] Roll fuchsia starnix: Avoid VMAR walks in procfs stats for traced_probes

Zircon VMAR walks (ZX_INFO_VMAR_MAPS) executed during
MemoryManager::get_stats() are extremely slow on large, fragmented JVM
processes like system_server. When Perfetto sweeps task status files,
this latency triggers the kTaskRunnerHung watchdog.

To fix this, we return a 4KB (1 page) constant stub (stub_memory_stats())
for Vm/Rss fields inside /proc/<pid>/stat, /proc/<pid>/statm, and
/proc/<pid>/status, but ONLY when the reader is `traced_probes`. This
cuts status file read latency during trace sweeps from 100ms+ down to
<1ms.

By limiting this workaround to `traced_probes`, we ensure that:
1. Other system daemons like `lmkd` still get accurate memory stats.
2. Standard procfs conformance tests (like gVisor`s
   `ProcSelfStat.PopulateWriteRSS` which expects RSS to grow upon
   allocation) continue to pass successfully.

A permanent fix requires an optimized Zircon memory aggregation API
(b/525059309).

Original-Original-Bug: 460841705
Original-Original-Bug: 525059309
Test: Verified on device that traced_probes sweeps succeed.
Test: ProcSelfStat.PopulateWriteRSS passes.
Original-Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/1683037
Original-Original-Revision: 804982ca4a3d30ebc12c268492b95a80e19e9b1b
GitOrigin-RevId: cf7ad17e78df02a76678b2bea4aa07f4f68b341b
Change-Id: Iea7377f5615b93cef9ab3dbfc8b6e1252b4bacfb
Reviewed-on: https://fuchsia-review.googlesource.com/c/integration/+/1685258
Cr-Commit-Position: refs/heads/main@{#195802}