Return diagnostics snapshot on first Stats and refactor diagnostics collection

[mateeullahmalik]

Motivation

• Initial Stats calls could return an empty/zeroed diagnostics snapshot because the expensive diagnostics refresh was only scheduled async, causing metrics to be missing on first call.
• Ensure the first Stats invocation can synchronously obtain and return a populated diagnostics snapshot so status consumers get meaningful data immediately.
• Reuse the same diagnostics collection logic for both sync and async refresh paths to avoid duplication and racey cache propagation.
• Improve observability by logging failures and slow refreshes for initial and background refreshes.

Description

• Stats now calls refreshDiagnosticsSync and uses the returned *StatsSnapshot when the cache is empty instead of relying on cache propagation.
• Introduced collectDiagnostics(ctx,p,prev) (replaces refreshDiagnostics) which returns (*StatsSnapshot, error) and performs the heavy snapshot collection without mutating global state directly.
• refreshDiagnosticsSync and maybeRefreshDiagnostics now both call collectDiagnostics and, when a snapshot is returned, call m.setSnapshot(next) and m.markFresh(); both paths remain deduped by refreshInFlight.
• Added warning logs for initial refresh failures and for refreshes slower than p2pStatsSlowRefreshThreshold to aid debugging.

Testing

• No automated tests were run.
• Changes were applied to p2p/p2p_stats.go and committed.

---

Codex Task

[roomote-v0]

Rooviewer   See task on Roo Cloud

Review complete. No issues found.

The changes correctly address the motivation of ensuring the first Stats call returns a populated diagnostics snapshot. The refactoring cleanly separates collection logic from caching, and the concurrency handling with refreshInFlight is sound.

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}