chore: Lint all cargo targets in CI

[jsonbailey]

Summary

CI already runs clippy with -D warnings, so lint findings do fail the build. What it does not do is pass --all-targets, so only the library target is linted — test modules and examples are never checked, and findings there accumulate invisibly.

Adding --all-targets surfaced four, fixed here:

Finding	Location
useless_borrows_in_formatting	src/events/sender.rs:356
unused import MultiContextBuilder	src/client.rs:924
unused imports ContextBuilder, mockito::Matcher, test_case::test_case, ServiceEndpointsBuilder, EventFactory	src/events/processor_builders.rs:342-347
example fails to compile, HyperTransport::new_https not found (E0599)	examples/custom_transport.rs:67

The unused imports are not dead code. They serve tests behind #[cfg(any(feature = ...))] gates, and the SDK's default feature set satisfies those gates, so the imports only go unused under --no-default-features variants. Each import is now gated with the same expression as the test that consumes it rather than deleted. maplit::hashset is deliberately left ungated — the test using it has no gate.

The example was an oversight rather than a lint problem. print_flags and progress each declare required-features = ["hyper-rustls-native-roots"], but custom_transport had no [[example]] entry at all, so cargo built it unconditionally even though new_https() needs rustls. It now declares the same requirement as its siblings; the example's code is unchanged. This was latent because the two matrix entries that would have broken it also pass cargo-test-flags: --lib, so nothing compiled the example there.

Testing

cargo clippy --all-targets -p launchdarkly-server-sdk -- -D warnings is clean for all seven CI matrix feature combinations, asserted on exit code rather than on output:

Flags	Result
(default)	clean
--no-default-features	clean
--no-default-features --features hyper	clean
--no-default-features --features hyper-rustls-native-roots	clean
--no-default-features --features hyper-rustls-webpki-roots	clean
--no-default-features --features native-tls	clean
--no-default-features --features hyper-rustls-native-roots,crypto-openssl	clean

cargo test: 413 unit passed, 19 doc-tests passed, 0 failed — unchanged from main. The diff touches no #[test], #[tokio::test], or #[test_case] line, so no test is newly gated out. Unit tests also pass under three --no-default-features variants (397 / 408 / 408), and those counts tracking the feature gates is the cross-check that each gate matches its tests. cargo fmt --check clean. contract-tests was already clean under --all-targets and still is.

A consequence worth knowing

CI pins the toolchain from .github/variables/rust-versions.env (currently target=1.96), and check-rust-versions.yml opens an MSRV bump PR nightly. New toolchains add new lints, so widening clippy's scope means a future bump PR can go red on code nobody touched. That is the point — drift surfaces on the bump PR instead of hiding — but whoever handles those bot PRs should expect it. The sender.rs finding is an example: it does not fire on 1.96 and would have appeared at the next bump regardless.

Note this PR's own CI runs clippy on 1.96, so it verifies the --all-targets change but cannot confirm the 1.98-only finding is gone. That was verified locally.

Not in scope

--all-features does not compile, on an unrelated dependency conflict in hyper-http-proxy. Left alone.