Fix OnEventf argument forwarding in the composite log

[Robin1987China]

Wrapped logs receive malformed event messages from a composite log. compositeLog.OnEventf forwarded
its variadic arguments as a single []interface{} value instead of expanding them, so every format
verb in the format string was rendered against that one slice.

Before

log/composite with two wrapped logs, l.OnEventf("session %s seq %d", "FIX.4.4", 7):

expected: []string{"session FIX.4.4 seq 7"}
actual  : []string{"session [FIX.4.4 %!s(int=7)] seq %!d(MISSING)"}

Even a single argument is mangled (%s renders the slice itself):

expected: []string{"session FIX.4.4"}
actual  : []string{"session [FIX.4.4]"}

There is no error — the garbage is written to the log and only shows up on inspection.

Fix

One line in log/composite/composite_log.go:

-		log.OnEventf(format, a)
+		log.OnEventf(format, a...)

quickfix.Log.OnEventf is OnEventf(string, ...interface{}), so a was passed as one argument whose
dynamic type is a slice.

Affected paths

Any call that goes through a composite log. 29 OnEventf call sites pass arguments — in acceptor.go,
session.go, in_session.go, session_state.go, initiator.go, logon_state.go, latent_state.go,
not_session_time.go — and 13 of them pass two or more. OnIncoming / OnOutgoing / OnEvent are
non-variadic and are not affected; this is the only occurrence in the repository.

Test

TestCompositeLogOnEventfExpandsFormatArgs in log/composite/composite_log_test.go, with three subtests:

• single string arg — "session %s": shows even one argument was corrupted.
• mixed args — "session %s seq %d": the common case, and where %!d(MISSING) appears.
• slice arg — "tags %v count %d" with a []int: guards against a "join the args into a string and
  forward that" fix, which would corrupt the output for slice arguments.

Each asserts both wrapped logs receive the expanded result.

It is a standalone Test* function rather than a case in the existing CompositeLogTestSuite: that
suite SkipNow()s as a whole when MONGODB_TEST_CXN is unset, so a test added there would silently
never run without MongoDB. The new test has no external dependencies — no MongoDB, no SQL, no network,
no credentials.

Verification

• go build ./... — clean
• go test -run TestCompositeLogOnEventfExpandsFormatArgs -v ./log/composite/ — 3/3 PASS
• go test -race ./log/composite/ — ok
• gofmt -l . and gofmt -s -l . — no output
• make vet — exit 0
• make test-ci — all packages pass except log/mongo, which fails identically on the unmodified
  baseline (see below); log/composite passes
• golangci-lint (v1.64.6, repo .golangci.yml, GOTOOLCHAIN=go1.23.12) — 0 issues, same as baseline

Repeating the test with the one line reverted turns all three subtests red again, so the fix is what
drives them.

What I could not verify

• No MongoDB locally, so make test (it hardcodes MONGODB_TEST_CXN=mongodb://db:27017) was not run;
  I ran the CI job's command make test-ci instead. log/mongo fails there both with and without my
  change — without a connection string the suite skips, then panics on a nil dereference — which is
  already tracked by Test: resolve panic when MONGODB_TEST_CXN is not set #755. Not caused by, and not fixed by, this PR.
• CI lints on Go 1.21; locally only go1.26.5 is available. The 0-issue lint result was obtained on
  1.23.12 (the version declared in go.mod), not on 1.21.
• make lint does not work locally as written (the Makefile installs the binary into ./bin but calls
  bare golangci-lint); I did not touch it.
• make accept was not run — it needs ruby, make generate and an echo server. This change does not
  touch parsing, protocol or generated code.
• No end-to-end session against a real counterparty was observed. The evidence is the unit test plus
  the call sites listed above.