fix(storage): widen the Windows ADS query timeout for cold Add-Type compiles

[liuxiaocs7]

Summary

The managed-dependency alternate-stream guard (assertNoWindowsAlternateStreams in packages/storage/src/managed-dependency-environment.ts) spawns a fresh powershell.exe per query that JIT-compiles the FindFirstStreamW P/Invoke helper via Add-Type. On a cold GitHub-hosted Windows runner that first compile — a cold csc.exe launch plus .NET Framework warmup — was observed at ~31s, tripping the 30s WINDOWS_STREAM_QUERY_TIMEOUT_MS. It then rejected with Timed out querying Windows alternate data streams, which is the wrong error for the test's assert.rejects, so the "rejects an NTFS alternate stream created inside a dependency artifact" subtest failed. The next two subtests reused the now-warm compiler and passed in 2–4s.

This flaked the unfiltered main-push Windows recovery lane (run 33096120938) even though the triggering commit only touched desktop tests. The lane had been green for the prior ~4 hours; the ADS code itself was unchanged since #3789.

The tree walk feeding the query is bounded and non-recursive, so the query cannot hang — the timeout exists only to bound a wedged interpreter. This raises it to 120s so the one-time cold compile has generous headroom rather than being misreported as a wedge. The 45-minute job budget still bounds a true hang. I kept the P/Invoke design (which deliberately replaced the recursive Get-Item -Stream walk) rather than rewrite a security boundary; the cold compile is the sole ~30s cost.

Refs #3789

Verification

• Change is a single constant widening (30_000 → 120_000) plus an explanatory comment; no behavioral change to the query or its callers.
• No deterministic regression test accompanies it: the failure is a runner-cold-start timing artifact, not a logic bug, so there is no input that fails at 30s but passes at 120s on a warm machine. The existing three Windows ADS subtests continue to assert correctness; this only removes their cold-start flake.
• Note on local checks: tsc/suite runs could not be executed in my environment because dependencies are not installed (@types/node absent). CI (test + Windows recovery) will exercise the build and the affected Windows suite.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code diagnosed the failing run and authored the one-line timeout change and its comment. Commit carries a Generated-by: Claude Code trailer.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

LGTM.