Replace C-style casts with static_cast/reinterpret_cast for -Wold-style-cast compliance

[PavelGuzenfeld]

Summary

Fixes #444

Replace all C-style casts with static_cast / reinterpret_cast across the three public headers, enabling compilation with -Wold-style-cast -Wuseless-cast -Werror on GCC 14/15.

• lightweightsemaphore.h (5 casts): (time_t) → static_cast<time_t>, (long) → static_cast<long>, (std::uint64_t) → static_cast<std::uint64_t>, (int) → static_cast<int>
• blockingconcurrentqueue.h (15 casts): All (size_t), (ssize_t), (int), (LightweightSemaphore::ssize_t) → static_cast<>. Assert (BlockingConcurrentQueue*)1 → reinterpret_cast<>
• concurrentqueue.h (1 cast): Suppress -Wuseless-cast around hash_thread_id — the static_cast<size_t> is needed for 32-bit platforms but redundant on 64-bit

Test plan

• GCC 15.2.0 with -std=c++23 -Wall -Wextra -Wpedantic -Werror -Wold-style-cast -Wuseless-cast -Wconversion -Wsign-conversion — clean
• GCC 14.2.0 with same flags — clean
• C++17, C++20, C++23 — all clean
• No behavioral changes — purely mechanical cast replacements

Test results

All tests executed inside Docker (gcc:latest, GCC 15.2.0):

Unit tests (make tests && bin/unittests)

All tests passed:

• create_empty_queue, enqueue_one, enqueue_many, enqueue_and_dequeue, enqueue_dequeue_threaded
• token_move, blocking_wrappers, c_api_create, c_api_enqueue, c_api_destroy
• acquire_and_signal, try_acquire_and_signal, core_add_only_list, core_thread_local
• core_free_list, core_spmc_hash, explicit_strings_threaded, large_traits

Fuzz tests (bin/fuzztests)

700+ iterations executed, 0 failures:

Test	Passed	Failed
multithread_produce	122	0
multithread_consume	114	0
multithread_produce_and_consume	132	0
completely_random	110	0
core_add_only_list	102	0
core_thread_local	124	0

[cameron314]

Personally, I think the code is less readable when using static_cast for integral conversions. However, it's probably more important to avoid warnings.

[cameron314]

This is not equivalent when sizeof(BlockingConcurrentQueue::ssize_t) != sizeof(LightweightSemaphore::ssize_t).

If (ssize_t)count is not negative (and it shouldn't be here!) then this works. But I'd still prefer an assert rather than an implicit assumption.

[PavelGuzenfeld]

I've restored the two-step cast through ssize_t and added assert(static_cast<ssize_t>(count) >= 0) before each conversion. Applied consistently across all 10 call sites in the file.