Acquire EB closure age

[coot]

Description

Added NominalDiffTime to TraceLeiosBlockTxsAcquired. This will
allow us to compute 90, 95 and 99 percentiles of how long it takes to acquire
all txs announced in an EB.

WARNING

To update your feature branch if it's stale, please rebase it manually on top of main. Don't update your feature branch by merging main into it. Your pull request will not pass CI if you do.

[ch1bo]

Getting ahold of the immutable tip each time here just for the sake of logging feels weird. Good enough for a prototype, but we probably could just figure out the wall clock time of the announcement once (we need to judge it coming late anyways) and just take the diff to that here.

[coot]

I asked claude:

This is already a known gap, not solved anywhere: InFutureCheck.hs:147-151's judgeHeaderArrival has the identical pattern — recompute
hardForkSummary/runQuery on every header — with a literal -- TODO cache this in the KnownIntersectionState? Or even in the LedgerDB? sitting unresolved. I
checked for a shared "cached interpreter" utility (mkInterpreter call sites, GetInterpreter handling) — the only other place that builds one is the
local-state-query handler (HardFork/Combinator/Ledger/Query.hs:530), which is also on-demand, not cached. So there's nothing to reuse; we'd be writing the
first instance of this pattern.

This can be fixed locally in newSlotAge by reusing the interpreter until it fails, and then caching it again, e.g.

  newSlotAge ::
    (IOLike m, HasHardForkHistory blk) =>
    SystemTime m ->
    TopLevelConfig blk ->
    m (ExtLedgerState blk EmptyMK) ->
    m (SlotNo -> m NominalDiffTime)
  newSlotAge systemTime cfg getLedger = do
    cacheVar <- StrictSTM.newTVarIO =<< mkInterp
    pure $ \slot -> do
      now <- systemTimeCurrent systemTime
      interp <- StrictSTM.readTVarIO cacheVar
      onset <- case interpretQuery interp (HardFork.slotToWallclock slot) of
        Right (onset, _slotLength) -> pure onset
        Left _pastHorizon -> do
          interp' <- mkInterp
          StrictSTM.atomically $ StrictSTM.writeTVar cacheVar interp'
          fst <$> HardFork.runQueryThrow (HardFork.slotToWallclock slot) -- via interp', see note below
      pure $ now `diffRelTime` onset
   where
    mkInterp = mkInterpreter . hardForkSummary (configLedger cfg) . ledgerState <$> getLedger

@nfrisby what do think?

[coot]

I feel it doesn't fit into this PR to fix the general problem, but I could do a separate one.

[ch1bo]

This is not what I worried about, neither what I suggested.

we probably could just figure out the wall clock time of the announcement once

Was meant to move the slot to wall clock time conversion into where we process (or create) announcements. Then, all the dependent timings we want to trace (when we acquired the corresponding EB body, closure, when we voted on it etc.) would just be relative to that point in time (ideally using a monotonic clock).

Either way, no change needed in this PR.

[ch1bo]

Seems like @nfrisby did the same in his #2237 work in 831a3c0