[Fix] Record confirmed transmission latency regardless of age

[thep2p]

Motivation

add_transmission_latency_metric checks staleness before checking whether a transmission was included in the block. Because a tracker entry lingers from first-seen until it's either confirmed or ages out, any transmission that confirms after the 30-minute threshold takes the stale branch:

if elapsed > 30min  → count STALE_UNCONFIRMED, drop     ← runs first                                                                                                                                                                                                                                                                                      
else if in block    → record TRANSMISSION_LATENCY, drop ← never reached when old 

Two consequences, both wrong:

• TRANSMISSION_LATENCY is silently truncated at 30 min — the slowest confirmations, exactly the tail you watch, never reach the histogram, biasing p99/max low.
• STALE_UNCONFIRMED_{SOLUTIONS,TRANSACTIONS} is polluted with transmissions that did land in the block.

It only surfaces under congestion / post-stall recovery (healthy-network confirmations are seconds), but those are precisely the moments the metric is supposed to be trustworthy.

Test Plan

No unit test: the function is pure side-effects on the global metrics recorder (untestable without a recorder harness), and the post-fix precedence is self-evident — adding an abstraction solely to make it testable would bury a one-branch semantic fix.

Documentation

Observability-only — no consensus, liveness, or funds impact.

Backwards compatibility

None affected — metric semantics only; no wire, storage, or protocol changes.