Use pod_vector for mmap zero-copy support in serialized data structures

[adamant-pwn]

Summary

This PR replaces std::vector with essentials::pod_vector in all serialized data structure members across bits. This is a companion to jermp/essentials#11, which introduces pod_vector and mmap zero-copy loading support in generic_loader.

No behavioral change for existing code — pod_vector is a drop-in replacement for std::vector when used in owned mode (the default). The change enables downstream applications to use generic_loader::set_mmap() to deserialize these data structures via zero-copy views into memory-mapped files. See the essentials PR for full motivation, benchmarks, and design details.

Changes

File	Member	Before	After
bit_vector.hpp	m_bits	std::vector<uint64_t>	pod_vector<uint64_t>
compact_vector.hpp	m_bits	std::vector<uint64_t>	pod_vector<uint64_t>
rank9.hpp	m_block_rank_pairs	std::vector<uint64_t>	pod_vector<uint64_t>
darray.hpp	m_positions	std::vector<uint64_t>	pod_vector<uint64_t>
cache_line_elias_fano.hpp	m_data	std::vector<uint8_t>	pod_vector<uint8_t>
endpoints_sequence.hpp	m_hints_0	std::vector<uint8_t>	pod_vector<uint8_t>

Additionally, darray.hpp templatizes access() and at() to accept both const uint64_t* and pod_vector<uint64_t>::const_iterator for const correctness.

Depends on: jermp/essentials#11

[adamant-pwn]

Should I update submodule in the PR to point to jermp/essentials@e2bd442?

[jermp]

Yes, thank you!

[adamant-pwn]

Alright, should be mergeable now 🙂

[jermp]

Done! thank you again :)