feat(io_uring): Add support for registered buffers

[kavirajk]

Changes

Added following new apis to the scheduler

1. registerBuffers()
2. readFixed()
3. writeFixed()

Which internally using liburing helpers io_uring_register_buffers(), io_uring_prep_read_fixed(), io_uring_prep_write_fixed().

This let us register the pre-allocated buffers that iouring can use during IO operations rather then allocating it per-io.

This is mainly based on best practices learned from TUM DBMS paper
https://arxiv.org/pdf/2512.04859

I also integrated with file-perf benchmark. And the numbers looked promising. See below for the actual improvement numbers

Performance

My setup is m8id.8xlarge EC2 instance

Normal

ec2-dev$ ./bb -b release perf --duration 60s --warmup 10s file
2026-06-12 21:48:21.567 [INFO ] bb:1986: command=perf preset=release
[0/2] Re-checking globbed directories...
ninja: no work to do.

## file-perf -- async file I/O

file=/dev/shm/file-perf.bin, bs=4k, size=1g, duration=60s, warmup=10s

| numjobs  | iodepth  | mode       | IOPS     | BW         | avg      | p50      | p95      | p99      | p99.9    |
|----------|----------|------------|----------|------------|----------|----------|----------|----------|----------|
| 1        | 1        | randwrite  | 185k     | 721.0 MiB/s | 5.39 µs  | 3.06 µs  | 12.83 µs | 13.57 µs | 22.39 µs |
| 1        | 16       | randwrite  | 575k     | 2245.0 MiB/s | 27.81 µs | 26.82 µs | 36.7 µs  | 41.02 µs | 54.04 µs |
| 16       | 1        | randwrite  | 893k     | 3489.0 MiB/s | 17.89 µs | 18.79 µs | 26.6 µs  | 37.29 µs | 52.2 µs  |
| 16       | 16       | randwrite  | 805k     | 3143.0 MiB/s | 318.16 µs | 262.8 µs | 850.48 µs | 1333.68 µs | 1847.18 µs |
| 1        | 1        | randread   | 232k     | 906.0 MiB/s | 4.29 µs  | 2.42 µs  | 12.35 µs | 12.99 µs | 20.79 µs |
| 1        | 16       | randread   | 682k     | 2665.0 MiB/s | 23.43 µs | 25.55 µs | 29.8 µs  | 38.12 µs | 53.12 µs |
| 16       | 1        | randread   | 2663k    | 10404.0 MiB/s | 5.98 µs  | 3.91 µs  | 15.24 µs | 29.43 µs | 98.45 µs |
| 16       | 16       | randread   | 4955k    | 19355.0 MiB/s | 51.63 µs | 50.92 µs | 80.68 µs | 99.89 µs | 126.67 µs |

Fixed Buffers

ec2-dev$ ./bb -b release perf --duration 60s --warmup 10s file --fixed-buffers
2026-06-12 21:58:31.006 [INFO ] bb:1986: command=perf preset=release
[0/2] Re-checking globbed directories...
ninja: no work to do.

## file-perf -- async file I/O

file=/dev/shm/file-perf.bin, bs=4k, size=1g, duration=60s, warmup=10s

| numjobs  | iodepth  | mode       | IOPS     | BW         | avg      | p50      | p95      | p99      | p99.9    |
|----------|----------|------------|----------|------------|----------|----------|----------|----------|----------|
| 1        | 1        | randwrite  | 193k     | 754.0 MiB/s | 5.15 µs  | 2.79 µs  | 12.83 µs | 13.65 µs | 22.48 µs |
| 1        | 16       | randwrite  | 608k     | 2373.0 MiB/s | 26.31 µs | 25.26 µs | 34.84 µs | 40.16 µs | 53.23 µs |
| 16       | 1        | randwrite  | 1117k    | 4362.0 MiB/s | 14.3 µs  | 14.89 µs | 23.68 µs | 29.79 µs | 39.66 µs |
| 16       | 16       | randwrite  | 1040k    | 4063.0 MiB/s | 246.11 µs | 222.31 µs | 505.13 µs | 901.61 µs | 1277.51 µs |
| 1        | 1        | randread   | 236k     | 924.0 MiB/s | 4.21 µs  | 2.38 µs  | 12.33 µs | 12.97 µs | 20.76 µs |
| 1        | 16       | randread   | 694k     | 2711.0 MiB/s | 23.03 µs | 25.11 µs | 29.24 µs | 37.22 µs | 52.57 µs |
| 16       | 1        | randread   | 2710k    | 10587.0 MiB/s | 5.88 µs  | 3.8 µs   | 15.16 µs | 28.89 µs | 100.26 µs |
| 16       | 16       | randread   | 5495k    | 21465.0 MiB/s | 46.56 µs | 45.05 µs | 71.7 µs  | 90.96 µs | 120.31 µs |

Throughput diff (normal vs fixed buffers)

| numjobs | iodepth | mode | IOPS before | IOPS after | Δ | BW before | BW after | Δ |
|---------|---------|-----------|-------------|------------|---------|-----------|----------|---------|
| 1       | 1       | randwrite | 185k        | 193k       | +4.3%   | 721.0     | 754.0    | +4.6%   |
| 1       | 16      | randwrite | 575k        | 608k       | +5.7%   | 2245.0    | 2373.0   | +5.7%   |
| 16      | 1       | randwrite | 893k        | 1117k      | +25.1%  | 3489.0    | 4362.0   | +25.0%  |
| 16      | 16      | randwrite | 805k        | 1040k      | +29.2%  | 3143.0    | 4063.0   | +29.3%  |
| 1       | 1       | randread  | 232k        | 236k       | +1.7%   | 906.0     | 924.0    | +2.0%   |
| 1       | 16      | randread  | 682k        | 694k       | +1.8%   | 2665.0    | 2711.0   | +1.7%   |
| 16      | 1       | randread  | 2663k       | 2710k      | +1.8%   | 10404.0   | 10587.0  | +1.8%   |
| 16      | 16      | randread  | 4955k       | 5495k      | +10.9%  | 19355.0   | 21465.0  | +10.9%  |

Latency diff (normal vs fixed buffers)

| numjobs | iodepth | mode      | avg                     | p50                     | p95                      | p99                       | p99.9                     |
|---------|---------|-----------|-------------------------|-------------------------|--------------------------|---------------------------|---------------------------|
| 1       | 1       | randwrite | 5.39→5.15 (−4.5%)       | 3.06→2.79 (−8.8%)       | 12.83→12.83 (0.0%)       | 13.57→13.65 (+0.6%)       | 22.39→22.48 (+0.4%)       |
| 1       | 16      | randwrite | 27.81→26.31 (−5.4%)     | 26.82→25.26 (−5.8%)     | 36.7→34.84 (−5.1%)       | 41.02→40.16 (−2.1%)       | 54.04→53.23 (−1.5%)       |
| 16      | 1       | randwrite | 17.89→14.3 (−20.1%)     | 18.79→14.89 (−20.8%)    | 26.6→23.68 (−11.0%)      | 37.29→29.79 (−20.1%)      | 52.2→39.66 (−24.0%)       |
| 16      | 16      | randwrite | 318.16→246.11 (−22.6%)  | 262.8→222.31 (−15.4%)   | 850.48→505.13 (−40.6%)   | 1333.68→901.61 (−32.4%)   | 1847.18→1277.51 (−30.8%)  |
| 1       | 1       | randread  | 4.29→4.21 (−1.9%)       | 2.42→2.38 (−1.7%)       | 12.35→12.33 (−0.2%)      | 12.99→12.97 (−0.2%)       | 20.79→20.76 (−0.1%)       |
| 1       | 16      | randread  | 23.43→23.03 (−1.7%)     | 25.55→25.11 (−1.7%)     | 29.8→29.24 (−1.9%)       | 38.12→37.22 (−2.4%)       | 53.12→52.57 (−1.0%)       |
| 16      | 1       | randread  | 5.98→5.88 (−1.7%)       | 3.91→3.8 (−2.8%)        | 15.24→15.16 (−0.5%)      | 29.43→28.89 (−1.8%)       | 98.45→100.26 (+1.8%)      |
| 16      | 16      | randread  | 51.63→46.56 (−9.8%)     | 50.92→45.05 (−11.5%)    | 80.68→71.7 (−11.1%)      | 99.89→90.96 (−8.9%)       | 126.67→120.31 (−5.0%)     |