v2.1

[Thermadiag]

The version 2.1 introduced additional breaking changes:

• The modules charconv and format are now deprecated. They are still available, but moved to the legacy folder.
  Indeed, while I really like these, they do not belong to a library about containers. They will eventually be moved to another library, and removed from seq.
• The tagged_pointer.hpp file have been moved to the internal folder (private API).

Additional changes:

• The hashing framework has a better handling of transparent keys.
• The hashing framework now supports std::chrono::time_point and std::chrono::duration.
• The seq::net_sort algorithm has been slightly optimized/refactored and moved to net_sort.hpp header.
• The radix_set/map lower_bound() method has been corrected (compilation error).
• radix_set/map now supports std::chrono::time_point and std::chrono::duration as key.
• radix_set/map no longer have typedef prefix_iterator and prefix_const_iterator. Instead, member prefix_range() now returns a std::pair of regular iterators.
• All radix-based containers (seq::radix_set/map, seq::radix_hash_set/map) have been internally simplified.
• The seq::hold_any class has been refactored and simplified. A seq::hold_any containing a char* or const char* is now considered holding a string. Comparison of seq::hold_any containing a char* or const char* results in string comparison.
• New container: seq::concurrent_queue.

The markdown documentation has been updated accordingly, and new benchmark results were added. Full list of benchmarks now:

• Benchmark of concurrent hash tables at the end of the page. Its goal is to compare seq::concurrent_set/map to other implementations.
• Benchmark on sorted containers. Its goal is to compare seq::flat_set/map and seq::radix_set/map to other implementations.
• Very tiny benchmark on concurrent queues to compare seq::concurrent_queue with other implementations.
• Memory and latency benchmark on hash tables to compare seq::radix_hash_set/map with other implementations.