clang 4 sanitizer is not happy with TBB usage in updater

[oxidase]

The travis clang 4.0 job with enabled sanitezer fails randomly in contraction or customization step
https://travis-ci.org/Project-OSRM/osrm-backend/jobs/212017889#L3688-L3696
ASAN sees the problem in leaked resources for TBB tasks like https://github.com/Project-OSRM/osrm-backend/blob/master/src/updater/updater.cpp#L276-L284 when update_edge_weights is false

==12703==T9 TSDDtor
==12703==T14 TSDDtor
==12703==T9 exited
==12703==T14 exited
==12703==T10 TSDDtor
==12703==T8 TSDDtor
==12703==T12 TSDDtor
==12703==T10 exited
==12703==T8 exited
==12703==T12 exited
==12703==T13 TSDDtor
==12703==T13 exited
==12703==T11 TSDDtor
==12703==T5 TSDDtor
==12703==T5 exited
==12703==T11 exited
==12703==T15: stack [0x7fb11ede2000,0x7fb11f1e1f80) size 0x3fff80; local=0x7fb11f1e1e88
==12703==T4 TSDDtor
==12703==T1 TSDDtor
==12703==T2 TSDDtor
[STXXL-ERRMSG] Removing disk file: /var/tmp/stxxl
==12703==T4 exited
==12703==T1 exited
==12703==T6 TSDDtor
==12703==T7 TSDDtor
==12703==T3 TSDDtor
ASAN:DEADLYSIGNAL
==12703==T2 exited
=================================================================
==12703==T7 exited
==12703==T3 exited
==12703==ERROR: AddressSanitizer: SEGV on unknown address 0xfffffffffffffff7 (pc 0x7fb133f260ba bp 0x000000000000 sp 0x7fb11f1e1de0 T15)
==12703==The signal is caused by a READ memory access.
    #0 0x7fb133f260b9  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x230b9)
    #1 0x7fb133f2618a  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x2318a)
    #2 0x7fb133f27ce7  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x24ce7)
    #3 0x7fb133f26641  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x23641)
    #4 0x7fb133f214a9  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x1e4a9)
    #5 0x7fb133f1dad4  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x1aad4)
    #6 0x7fb133f1dd58  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x1ad58)
    #7 0x7fb1341476c9  (/lib/x86_64-linux-gnu/libpthread.so.0+0x76c9)
    #8 0x7fb132d8f0ae  (/lib/x86_64-linux-gnu/libc.so.6+0x1080ae)

AddressSanitizer can not provide additional info.
SUMMARY: AddressSanitizer: SEGV (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x230b9) 
Thread T15 created by T6 here:
    #0 0x5c4ddd  (/home/miha/mapbox/osrm-backend/builds/clang-4.0/osrm-contract+0x5c4ddd)
    #1 0x7fb133f1d98c  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x1a98c)
    #2 0x7fb12fbeb67f  (<unknown module>)

Thread T6 created by T3 here:
    #0 0x5c4ddd  (/home/miha/mapbox/osrm-backend/builds/clang-4.0/osrm-contract+0x5c4ddd)
    #1 0x7fb133f1d98c  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x1a98c)

Thread T3 created by T1 here:
    #0 0x5c4ddd  (/home/miha/mapbox/osrm-backend/builds/clang-4.0/osrm-contract+0x5c4ddd)
    #1 0x7fb133f1d98c  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x1a98c)

Thread T1 created by T0 here:
    #0 0x5c4ddd  (/home/miha/mapbox/osrm-backend/builds/clang-4.0/osrm-contract+0x5c4ddd)
    #1 0x7fb133f1d98c  (/usr/lib/x86_64-linux-gnu/libtbb.so.2+0x1a98c)

It is not critical, but one possibility to fix it is to move update_edge_weights outside of loaders and make own task list

if (update_edge_weights)
{
  // root task???
  tbb::task_list list;
  list.push_back( *new( tbb::task::allocate_child() ) /*load_internal_to_external_node_map*/ [](){} );
  list.push_back( *new( tbb::task::allocate_child() ) /*load_geometries*/ [](){} );
  set_ref_count(3);
  spawn_and_wait_for_all();
}

Issue #3788 will double number of updater runs, so the clang 4.0 job will be red with high probability for ~2000 runs

[oxidase]

Fixed in #3809 by ed143b5