Update dependencies

[Michael-Wigham]

What does this PR do?

• Bumps MSRV to 1.92
  • Applies compiler-suggested std API updates and style cleanups.
  • Modernises internals for newer dependency versions
    • Moves file descriptor handling to OwnedFd/BorrowedFd where appropriate for clearer ownership and safer close semantics.
• Updates crate dependencies across the workspace, including updating the liburing submodule to 2.13.
• Updates sysfs probing logic:
  • Stops assuming minor == 0 identifies the parent block device; this is incorrect for partitions and NVMe devices where valid block devices frequently have non-zero minors
  • Adds partition-safe queue discovery (walk-up to find a queue/directory).
  • Makes queue/io_poll optional (if files are missing - assume unsupported) to handle older kernels/drivers and avoid hard failures when sysfs files are absent.
  • Simplifies cached block property access and removes the “probe iopoll by performing an I/O” path from the reactor.
  • Fixes CI failures on partition-backed and virtual block devices where queue/* files are missing or located on the parent disk.
• spsc_queue changes:
  • Eliminate manual allocation and raw pointer arithmetic for buffer, replacing it with an owned slice.
  • Cleans up lookahead/limit handling and switches head/tail progression to wrapping arithmetic to ensure correctness across wrap boundaries.
  • The underlying queue algorithm remains unchanged; this is a safety and maintainability improvement rather than a semantic change.
• Cleans up examples and benchmarks:
  • Removes unused hyper timer glue that was generating warnings.
  • Updates the SPSC benchmark to pin producer/consumer threads to CPUs and prints clearer latency/throughput numbers.

Motivation

This is a maintenance PR to keep Glommio building cleanly on modern toolchains and dependency versions, and to unblock CI failures.

Keeping the crate aligned with recent stable Rust:

• Reduces friction for contributors using current toolchains
• Prevents accumulating breakage from upstream ecosystem changes
• Lowers the long-term maintenance cost for future feature work

Notes on behaviour changes

• Poll-ring eligibility is now determined by sysfs capability values (e.g. queue/io_poll) combined with conservative device gating, rather than probing support via a test DMA read.
  • This avoids partially-working states where a device appears pollable until the first I/O fails.
  • This may relate to DmaFile returns EOPNOTSUP when reading (error code 95) #605 (devices appearing “pollable” until the first read routes via the poll ring and fail).
• Sysfs reads are now more defensive:
  • Missing files fall back to safe defaults instead of panicking.
  • This ensures Glommio continues to work on older kernels, non-NVMe devices, and CI environments.

Related issues

No related issues - standalone MR.

Additional Notes

• No public API changes.
• Functional behavior is unchanged for NVMe devices with full sysfs support.
• Behaviour on non-NVMe, virtual, or partition-backed devices avoids runtime and CI failures.
• The examples workspace is now edition = "2024" to align with the updated toolchain and formatting expectations.
• The SPSC benchmark output is intended for easier interpretation in CI/local perf runs and is not a functional API change.
  • The new bench also supports comparing different core pairings; on SMT systems, this can highlight differences between sibling vs non-sibling cores.

Checklist

• [] I have added unit tests to the code I am submitting
• [] My unit tests cover both failure and success scenarios
• [] If applicable, I have discussed my architecture