[agent-p1p]

Problem

The "Check stable" CI job fails on master (de3031a) and on the open PRs #1494 and #1495. It builds with -D warnings, and Rust 1.99 deprecates AtomicUsize::fetch_update (renamed to try_update), which PendingReqCounter::decrement uses. Nobody using the crate is affected, but the failing job hides real failures on new PRs.

Fix

Replace the fetch_update call with the compare-and-swap loop it performs internally. try_update is newer than the 1.85 MSRV, so it can't be used yet, and writing out the loop avoids suppressing the lint. The decrement still saturates at zero.

Notes to the reviewers

• Checked locally with rustc 1.99.0, the version the failing job used. CARGO_BUILD_RUSTFLAGS="-D warnings" cargo +1.99.0 build --workspace --all-targets and cargo +1.99.0 clippy --all-targets -- -D warnings both pass. On master, the build fails with the same error as CI.
• The existing pending_req_counter_does_not_underflow and pending_req_counter_tracks_requests_until_all_are_dropped tests cover the behaviour. just precommit passes.
• No changelog entry, since there is no user-visible change.

Checklist

• I followed the contribution guidelines
• I updated the relevant CHANGELOG.md (if applicable)
• I understand and can explain all code in this PR

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 85.71429% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
nostr-sdk/src/relay/inner.rs	85.71%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[agent-p1p]

@yukibtc The loop is only needed because of the MSRV. try_update, the replacement, has been stable since Rust 1.95 (fetch_update was deprecated in 1.99), while the workspace MSRV is 1.85. If you're willing to raise the MSRV to 1.95, the cleaner fix is a one-line rename to try_update, and I'm happy to switch this PR to that. If not, this version keeps 1.85 working without silencing the lint.