Unify global umf_ba_pool_t initialization

[igchor]

Initializing the global pool in libumf and in each pool separately led to problems with double initialization when multiple pools were used in a single application (and static libumf).

To avoid this problem, use util_init_once inside umf_ba_get_pool instead of having explicit constructor/destructor.

This fixes an issue exposed by: 0c55d8f

[igchor]

This should have been tested by #223 but scalable pool was missing from the test. It's fixed here: #228

[ldorau]

LGTM

[ldorau]

@igchor There is an issue here. I run ubench with proxy library in LD_PRELOAD:

$ LD_PRELOAD=./lib/libumf_proxy.so ./benchmark/ubench

1. The constructor of libumf_proxy.so is called.
2. The constructor runs umfMemoryProviderCreate(umfOsMemoryProviderOps()) that creates the global base allocator no. 1
3. The constructor runs umfPoolCreate(umfJemallocPoolOps()) that creates the global base allocator no. 2 and allocates je_malloc pool inside it.
4. The benchmark runs and ends.
5. atexit(umf_ba_destroy_global); is called and the global base allocator no. 2 is destroyed.
6. The destructor of libumf_proxy.so is called.
7. The destructor of libumf_proxy.so runs umfPoolDestroy(pool) and it blows up, because the memory it uses has already been freed in 5)

[ldorau]

However I think it should be merged as is, because it fixes the failure of #227 and refined later.