perf(pm): add resolver pm wiring

[elrrrrrrr]

Summary

Resolver stack split from #3028 / #2948, stacked after #3036.

Adds the PM-side wiring needed by the resolver source stack:

• uses a resolver-specific manifest concurrency default for non-semver/npmjs mode;
• keeps the user-visible install/download concurrency default unchanged;
• updates CLI help to document the resolver default;
• moves manifest-store persistence onto the writer queue shape used by the source PR;
• removes the obsolete compact sync JSON helper after manifest-store no longer uses it;
• removes preload progress event handling from the PM logger and adjusts the pipeline receiver test;
• keeps project-cache/retry docs aligned with resolver-owned cache semantics.

No resolver scheduling logic lands in this PR; it only prepares PM-side integration for the upcoming mainloop PRs.

Size

• Diff against perf(pm): fan out registry http clients #3036: 9 files changed, 115 insertions(+), 113 deletions(-)

Validation

• cargo fmt
• cargo check -p utoo-pm
• cargo test -p utoo-pm util::user_config::tests::test_resolver_manifest_concurrency_raises_npmjs_default
• cargo test -p utoo-pm service::pipeline::receiver::tests::test_pipeline_receiver_filters_events
• cargo clippy --all-targets -- -D warnings --no-deps

pack-napi still warns locally because next.js is a symlink in this worktree; clippy exits successfully.

[gemini-code-assist]

Code Review

This pull request adjusts manifest fetch concurrency, increasing the default to 256 for non-semver resolvers while maintaining 64 for others. It also refactors manifest persistence by removing the bounded write queue and switching from streaming JSON serialization to in-memory buffering. Feedback highlights concerns regarding the removal of backpressure in the manifest writer, which could lead to unbounded memory growth, and the increased memory overhead caused by allocating full vectors for JSON serialization instead of streaming to disk.

[gemini-code-assist]

The switch from a bounded sync_channel to an unbounded channel removes the backpressure mechanism. The previous implementation used a capacity of 1024 and dropped writes when full, which prevented unbounded memory growth if disk I/O stalled. Since manifest persistence is described as 'opportunistic' (line 150), dropping writes is a safer strategy than risking an OOM during large-scale dependency resolution.

[gemini-code-assist]

Using serde_json::to_vec allocates a full Vec<u8> for every manifest before writing. The previous implementation used serde_json::to_writer with a BufWriter, which streamed the JSON directly to the file. For large manifests, this new approach increases memory pressure and allocation overhead, which seems contrary to the 'perf' goal of this PR.

[github-actions]

📊 pm-bench-phases · b91b56d · linux (ubuntu-latest)

Workflow run — ant-design

PMs: utoo (this branch) · utoo-npm (latest published) · bun (latest)

npmjs.org

p0_full_cold

PM	wall	±σ	user	sys	RSS	pgMinor
bun	9.22s	0.26s	10.43s	10.19s	712M	327.8K
utoo-next	7.95s	0.17s	10.50s	12.08s	944M	118.4K
utoo-npm	8.14s	0.02s	10.85s	12.19s	1010M	115.3K
utoo	8.52s	0.18s	11.03s	12.22s	980M	124.1K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	14.2K	18.7K	1.19G	6M	1.88G	1.76G	1M
utoo-next	104.4K	75.4K	1.16G	4M	1.71G	1.70G	2M
utoo-npm	114.8K	79.5K	1.16G	4M	1.71G	1.70G	2M
utoo	113.5K	61.3K	1.16G	5M	1.71G	1.70G	2M

p1_resolve

PM	wall	±σ	user	sys	RSS	pgMinor
bun	1.96s	0.10s	4.06s	0.98s	514M	157.3K
utoo-next	2.84s	0.13s	5.27s	1.63s	619M	84.4K
utoo-npm	3.01s	0.09s	5.54s	1.92s	609M	77.8K
utoo	3.17s	0.07s	5.86s	1.85s	624M	90.5K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	7.8K	4.4K	203M	3M	107M	-	1M
utoo-next	45.4K	66.4K	199M	2M	7M	3M	2M
utoo-npm	68.9K	87.4K	199M	2M	7M	3M	2M
utoo	51.1K	71.0K	201M	3M	7M	3M	2M

p3_cold_install

PM	wall	±σ	user	sys	RSS	pgMinor
bun	6.60s	0.11s	6.22s	9.82s	638M	205.8K
utoo-next	8.03s	2.20s	5.03s	11.16s	535M	61.2K
utoo-npm	5.99s	0.09s	5.02s	10.86s	473M	59.4K
utoo	5.96s	0.59s	5.01s	10.61s	484M	64.9K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	5.0K	6.9K	1018M	3M	1.76G	1.76G	1M
utoo-next	120.0K	50.1K	989M	3M	1.70G	1.70G	2M
utoo-npm	94.8K	46.2K	989M	2M	1.70G	1.70G	2M
utoo	88.4K	50.2K	989M	2M	1.70G	1.70G	2M

p4_warm_link

PM	wall	±σ	user	sys	RSS	pgMinor
bun	3.29s	0.05s	0.20s	2.36s	133M	32.7K
utoo-next	2.33s	0.05s	0.51s	3.80s	79M	18.6K
utoo-npm	2.38s	0.09s	0.49s	3.80s	79M	18.2K
utoo	2.30s	0.03s	0.49s	3.79s	78M	18.6K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	422	25	5M	16K	1.91G	1.75G	1M
utoo-next	44.7K	18.9K	3K	25K	1.70G	1.70G	2M
utoo-npm	42.1K	18.2K	2K	9K	1.70G	1.70G	2M
utoo	42.2K	20.0K	1K	5K	1.71G	1.70G	2M

npmmirror.com: no output captured.