Refine and commit all temporary metrics from DanShaders:ton:more-metrics

[avevad]

[github-actions]

• validator/manager.cpp:3851: update_block_receive_stats() increments blocks_new_total again whenever a new receive type is observed after the block was already marked applied. That makes new_total count late observations rather than blocks, so a block applied first and then later seen via download/broadcast/candidate can be counted multiple times. This path should either update received_from_total_ for the late type or skip counter changes, but it should not increment new_total_ again.

[github-actions]

• validator/manager.cpp: new_block_broadcast records the broadcast source before ValidateBroadcast succeeds. Invalid broadcasts can poison first_received_from_total/received_from_total for a block that is later applied. Record the source only after co_await task succeeds.

• validator/manager.cpp: new_total_ is incremented again for every new source observed after a block was already marked applied, even though the block was already counted in new_block_cont. This makes blocks_new_total count observations, not blocks, and late sources are not reflected in received_from_total_.

[github-actions]

• validator/manager.cpp:3850: update_block_receive_stats() increments new_total_ for any new receive type observed after stats.applied is already true. A duplicate/late broadcast or candidate for an already-applied block will therefore increase the total block counter again without incrementing received_from_total_ or first_received_from_total_, skewing the metric denominator. Late events should either be ignored for the applied-block totals or update the per-source counters without re-counting the block.