svn_fs_base__retry_txn() is a while(1) loop that calls begin_trail(pool) in every iteration. The latter function callocates a struct trail_t out of POOL.
Is there a memory growth issue here? Will the number of iterations of the while(1) loop be small or bounded? If not, should we avoid repeated allocations out of POOL? (e.g., by re-using a single trail_t struct, or by letting the struct live on stack until a successful txn)
