Fix: sessiond: snapshot: leak of trace chunk

Fix: sessiond: snapshot: leak of trace chunk

Valgrind reports a leak after every snapshot record command:

==827791== 430 (280 direct, 150 indirect) bytes in 1 blocks are definitely lost in loss record 34 of 37
==827791==    at 0x48435FF: calloc (vg_replace_malloc.c:1117)
==827791==    by 0x223D01: zmalloc (macros.h:45)
==827791==    by 0x224B79: lttng_trace_chunk_allocate (trace-chunk.c:387)
==827791==    by 0x224E41: lttng_trace_chunk_create (trace-chunk.c:427)
==827791==    by 0x150B55: session_create_new_trace_chunk (session.c:656)
==827791==    by 0x164A11: snapshot_record (cmd.c:5113)
==827791==    by 0x1651EE: cmd_snapshot_record (cmd.c:5302)
==827791==    by 0x196E74: process_client_msg (client.c:2166)
==827791==    by 0x198AF1: thread_manage_clients (client.c:2742)
==827791==    by 0x18E245: launch_thread (thread.c:66)
==827791==    by 0x4B9E258: start_thread (in /usr/lib/libpthread-2.33.so)
==827791==    by 0x4CB45E2: clone (in /usr/lib/libc-2.33.so)

session_set_trace_chunk() on line 5162 returns a reference to the
current trace chunk which is never released.

This also causes tests/regression/tools/snapshots/test_ust_long to fail
due to a file descriptor exhaustion (presumably from using too many
directory file descriptors) when it is executed by an unprivileged user.

The CI doesn't catch this since the long regression test suite is
executed as root.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I4b6e45df48c3daafa2294c80ccd8a2b4d91401e1