Problem: Each peer's send pause state is a flag plus a separate start timestamp. Message queueing and socket sends each duplicate the transition logic that keeps the two in sync, and the inactivity check (which needs both), reads them separately.
Fix: Move the pause transitions into CNode accessors over one atomic timestamp.
The inactivity check then reads the pause status and start time as one snapshot.
Follow-up to #36667.
The test send buffer limit is zero, so queued messages set `fPauseSend`.
Move the repeated resets into `ProcessMessagesOnce()`.
This adds resets at four call sites in `orphan_reconsideration_interruptible`, where the peer is already unpaused.
Keeping resets here will hide the later unpaused timestamp sentinel from test callers.
Hide the flag and timestamp behind `CNode` accessors and extract their shared updates into `SetPauseSend()`.
Both production update sites already hold `cs_vSend`.
Cover preserving an active pause's start time and recording a fresh one after resuming.
Encode the paused state in the timestamp, using epoch for unpaused.
This removes the separate flag and gives readers one atomic snapshot.
Preserve the original start time across repeated pause updates.
<!--e57a25ab6845829454e8d69fc972939a-->
The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.
<!--006a51241073e994b41acfe9ec718e94-->
<!--021abf342d371248e50ceaed478a90ca-->
See the guideline and AI policy for information on the review process. A summary of reviews will appear here.
<!--174a7506f384e20aa4161008e828411d-->
Reviewers, this pull request conflicts with the following ones:
clang-tidy optin.performance.Padding by fanquake)If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.
<!--5faf32d7da4f0f540f40219e4f7537a3-->