Compaction, retention, and soak testing

Block refcounting (atomic refs + condemned flag) lets live queries
survive concurrent compaction and retention. Levelled compactor merges
2h->8h->32h blocks via raw chunk passthrough. Retention drops blocks
older than the configured window. Background goroutine drives
flush/compact/retain cycles; exported RunCompaction/ApplyRetention allow
deterministic test control via injectable clock.

Soak test: 10k series x 48h simulated at 15s intervals (115M samples).
Validates flat memory (158 MiB peak), bounded disk (13 blocks), and zero
query errors during compaction.

All existing tests refactored to table-driven with uniform assertions.

internal/block/writer.go

@@ -30,6 +30,16 @@ type SeriesFlush struct {
 //
 // Returns the block ULID and any error.
 func Flush(dataDir string, series []SeriesFlush) (string, error) {
+	return flushBlock(dataDir, series, 1, nil)
+}
+
+// FlushCompacted writes a new immutable block from compacted series data,
+// recording the compaction level and source block ULIDs.
+func FlushCompacted(dataDir string, series []SeriesFlush, level int, sources []string) (string, error) {
+	return flushBlock(dataDir, series, level, sources)
+}
+
+func flushBlock(dataDir string, series []SeriesFlush, level int, sources []string) (string, error) {
 	ulid := newULID()
 	blockDir := filepath.Join(dataDir, ulid)
 
@@ -49,7 +59,12 @@ func Flush(dataDir string, series []SeriesFlush) (string, error) {
 	)
 	meta.ULID = ulid
 	meta.Version = 1
-	meta.Compaction = CompactionInfo{Level: 1, Sources: []string{ulid}}
+	meta.Compaction = CompactionInfo{Level: level}
+	if sources != nil {
+		meta.Compaction.Sources = sources
+	} else {
+		meta.Compaction.Sources = []string{ulid}
+	}
 	meta.MinTime = int64(^uint64(0) >> 1) // max int64
 	meta.MaxTime = int64(0)