[FEATURE] Use a sharded locking structure instead of a single lock for the Global Allocator

[KarthikSubbarao]

The problem/use-case that the feature addresses

Use a sharded locking structure instead of a single global lock.

There is a major bottle neck in fulltext (maybe even non vector queries) where the allocations from reader/writer threads are stuck waiting on the single global bottle neck. Depending on the workload, 30%-40% of time is just spent here.

An idea for reducing this bottle neck is to move from a single lock + hashset to a sharded system where have multiple ((e.g. 64 or 128)) such structures that track pointers and have their own individual locks

Example:

Samples: 423K of event 'task-clock:ppp', 4000 Hz, Event count (approx.): 72318823208 lost: 0/0 drop: 0/118909
Overhead  Shared Object         Symbol
  34.91%  libsearch.so          [.] absl::lts_20240722::Mutex::Lock()                                                                     ◆
   9.17%  libsearch.so          [.] valkey_search::indexes::text::RadixTree<valkey_search::indexes::text::InvasivePtr<valkey_search::index▒
   8.33%  libsearch.so          [.] vmsdk::ReportFreeMemorySize(unsigned long)                                                            ▒
   7.27%  libsearch.so          [.] vmsdk::ReportAllocMemorySize(unsigned long)                                                           ▒
   6.65%  libsearch.so          [.] __wrap_free                                                                                           ▒
   5.22%  libsearch.so          [.] absl::lts_20240722::Mutex::Unlock()                                                                   ▒
   3.05%  libsearch.so          [.] valkey_search::indexes::text::Postings::KeyIterator::ContainsFields(unsigned long) const              ▒
   2.57%  [kernel]              [k] 0xffffffff94cec011                                                                                    ▒
   1.90%  libsearch.so          [.] valkey_search::indexes::PrefilterEvaluator::EvaluateText(valkey_search::query::TextPredicate const&, b▒
   1.70%  libstdc++.so.6.0.33   [.] std::_Rb_tree_increment(std::_Rb_tree_node_base const*)                                               ▒
   1.42%  libsearch.so          [.] absl::lts_20240722::Mutex::LockSlowLoop(absl::lts_20240722::SynchWaitParams*, int)                    ▒
   1.36%  valkey-server         [.] VM_Alloc                                                                                              ▒
   1.13%  valkey-server         [.] valkey_free                                                                                           ▒
   0.92%  libc.so.6             [.] syscall                                                                                               ▒
   0.86%  libsearch.so          [.] absl::lts_20240722::Mutex::UnlockSlow(absl::lts_20240722::SynchWaitParams*)                           ▒
   0.80%  libsearch.so          [.] valkey_search::query::TermPredicate::Evaluate(valkey_search::indexes::text::TextIndex const&, valkey_s▒
   0.75%  [kernel]              [k] 0xffffffff93bd7ddd                                                                                    ▒
   0.75%  libsearch.so          [.] valkey_search::indexes::text::TermIterator::FindMinimumValidKey()                                     ▒
   0.70%  valkey-server         [.] je_malloc_usable_size                                                                                 ▒
   0.68%  libsearch.so          [.] valkey_search::query::TermPredicate::GetTextIndexSchema() const                                       ▒
   0.57%  libsearch.so          [.] valkey_search::indexes::text::Postings::KeyIterator::SkipForwardKey(valkey_search::InternedStringPtr c▒
   0.53%  libsearch.so          [.] std::deque<valkey_search::indexes::Neighbor, std::allocator<valkey_search::indexes::Neighbor> >::_M_de▒
   0.53%  [kernel]              [k] 0xffffffff93bd7df6                                                                                    ▒
   0.29%  [kernel]              [k] 0xffffffff94cd60d5                                                                                    ▒
   0.26%  libsearch.so          [.] absl::lts_20240722::Enqueue(absl::lts_20240722::base_internal::PerThreadSynch*, absl::lts_20240722::Sy▒
   0.23%  libsearch.so          [.] absl::lts_20240722::synchronization_internal::FutexWaiter::Wait(absl::lts_20240722::synchronization_in▒
   0.23%  [kernel]              [k] 0xffffffff93cc1f53                                                                                    ▒
   0.22%  libsearch.so          [.] absl::lts_20240722::synchronization_internal::FutexWaiter::Post()  

[KarthikSubbarao]

Addressed in fulltext here: #553