LedgerDB.V1: fix race between chain switch and forker acquisition (#1737)

This fixes a small bug introduced in #1619, which does not affect Praos.

### Explanation

Before this PR, it was possible that, when opening a forker, we first
would read the `DbChangelog`

https://github.com/IntersectMBO/ouroboros-consensus/blob/fd31cf8fd056d5ddee39c79400b3382d93bccb47/ouroboros-consensus/src/ouroboros-consensus/Ouroboros/Consensus/Storage/LedgerDB/V1.hs#L762
then switch to a different chain in

https://github.com/IntersectMBO/ouroboros-consensus/blob/fd31cf8fd056d5ddee39c79400b3382d93bccb47/ouroboros-consensus/src/ouroboros-consensus/Ouroboros/Consensus/Storage/ChainDB/Impl/ChainSel.hs#L915-L916
and only then read

https://github.com/IntersectMBO/ouroboros-consensus/blob/fd31cf8fd056d5ddee39c79400b3382d93bccb47/ouroboros-consensus/src/ouroboros-consensus/Ouroboros/Consensus/Storage/LedgerDB/V1.hs#L763

With Praos, this doesn't matter, because `getVolatileSuffix` will always
return the `k` most recent blocks/states, so there is no observable
difference. However, with Peras, it can be the case that the length of
the volatile suffix gets longer/shorter than before, leading to a
mismatch in the length of the volatile headers and volatile
`DbChangelog` states.

In practice, the only effect is that a user would erroneously not be
able to acquire points in the LocalStateQuery mini-protocol that it
should be able to acquire, or vice versa.

### Fix

The fix is to simply move these two things into the same `STM`
transaction, as is already the case in all other places in
LedgerDB.{V1,V2} that involve `ldbGetVolatileSuffix`.

A regression test would need to rely on IOSimPOR; therefore, I am
deferring this until we have sth like
#1494.

Author: amesgen

ouroboros-consensus/changelog.d/20251028_101139_alexander.esgen_ledgerdb_v1_race.md

@@ -0,0 +1,3 @@
+### Patch
+
+- Fix a race condition between chain switches and LedgerDB.V1 forker acquisition.

ouroboros-consensus/src/ouroboros-consensus/Ouroboros/Consensus/Storage/LedgerDB/V1.hs

@@ -759,13 +759,14 @@ acquireAtTarget ::
   Either Word64 (Target (Point blk)) ->
   ReadLocked m (Either GetForkerError (DbChangelog l))
 acquireAtTarget ldbEnv target = readLocked $ runExceptT $ do
-  dblog <- lift $ readTVarIO (ldbChangelog ldbEnv)
-  volSuffix <- lift $ atomically $ getVolatileSuffix $ ldbGetVolatileSuffix ldbEnv
-  -- The DbChangelog might contain more than k states if they have not yet
-  -- been garbage-collected.
-  let volStates = volSuffix $ changelogStates dblog
+  (dblog, volStates) <- lift $ atomically $ do
+    dblog <- readTVar (ldbChangelog ldbEnv)
+    -- The DbChangelog might contain more than k states if they have not yet
+    -- been garbage-collected.
+    volSuffix <- getVolatileSuffix $ ldbGetVolatileSuffix ldbEnv
+    pure (dblog, volSuffix $ changelogStates dblog)
 
-      immTip :: Point blk
+  let immTip :: Point blk
       immTip = castPoint $ getTip $ AS.anchor volStates
 
       rollbackMax :: Word64