chore: silence rust-analyzer's cyclic-deps log spam (#294)

Follow-up to f6e49c5. Opening the workspace floods the rust-analyzer
server log with "cyclic deps" warnings, one per dev-dependency cycle in
the crate graph.

They are all dev-dependency cycles, which Cargo allows. rust-analyzer
keeps one crate-graph node per package and does not separate dev-deps
from normal deps, so the standard "A dev-depends on B, B depends on A"
pattern always looks like a cycle to it. Nearly all of ours are upstream
reth's — reth-trie-db and reth-provider dev-depend on each other, and so
on. The two that touch our crates are reth-seismic-node's dev-dep on
itself, copied verbatim from reth-optimism-node, and
reth-seismic-primitives dev-depending on reth-seismic-test-utils for a
single test module. None of it is worth restructuring; upstream reth
configures nothing for these and lives with them.

So filter the log instead. RA_LOG is a tracing Targets filter, and
add_dep_inner is the only tracing::warn! in project_model::workspace, so
raising just that module to "error" drops these lines and keeps
everything else: the module's own tracing::error! calls still pass, and
every other module stays at "warn", rust-analyzer's own default.

Measured with the 1.95.0 rust-analyzer, running analysis-stats over
seismic-reth/crates/seismic/test-utils: 19 "cyclic deps" lines before, 0
after, and no other WARN or ERROR output in either run.

This does not address the separate task_pool.rs SendError panic on
server shutdown — that is a known upstream bug
(rust-lang/rust-analyzer#18055), unrelated to
logging config.

Author: samlaf

workspace/seismic.code-workspace

@@ -90,5 +90,22 @@
 		// up front costs several GB of memory per window, which is expensive in this workspace
 		// that contains so many different repos.
 		"rust-analyzer.cachePriming.enable": false,
+		// Silence the "cyclic deps" warnings that flood the server log. They are
+		// dev-dependency cycles, which Cargo allows: rust-analyzer keeps one node
+		// per package and does not separate dev-deps from normal deps, so the
+		// standard "A dev-depends on B, B depends on A" pattern always looks like a
+		// cycle to it. Most of ours come from upstream reth (reth-trie-db and
+		// reth-provider dev-depend on each other, and so on), so there is nothing
+		// to fix on our side.
+		//
+		// The warning is logged by add_dep_inner in
+		// https://github.com/rust-lang/rust-analyzer/blob/master/crates/project-model/src/workspace.rs
+		// which is the only tracing::warn! in that module — everything else there
+		// is tracing::error!, so raising just this module to "error" loses nothing.
+		// "warn" is rust-analyzer's own default filter, kept here for every other
+		// module. RA_LOG is a tracing Targets filter, hence the module::path syntax.
+		"rust-analyzer.server.extraEnv": {
+			"RA_LOG": "warn,project_model::workspace=error"
+		},
 	}
 }