Column can expire while lazy compacting it...

Using TTL + range tombstones can lead to failure while lazy compacting rows.

Scenario to reproduce :

• create an SSTable with one row and some columns and a TTL of 8 seconds
• wait one second
• create a second SSTable with the same rowkey as above, and add a range tombstone
• start the first pass of the lazy compaction before the columns with TTL are expired
• wait 10 seconds (enough for columns with TTL to expire)
• continue lazy expiration
• the following assertion will fail :
  [junit] junit.framework.AssertionFailedError: originally calculated column size of 1379 but now it is 1082
  [junit] at org.apache.cassandra.db.compaction.LazilyCompactedRow.write(LazilyCompactedRow.java:150)