ci: native-runner wheels for all 4 arches; fix sdist readme path

[Xof]

Summary

The Build wheels release workflow (wheels.yml) was failing on the
v0.1.0-exilis-catalog-dep tag. This fixes it, plus a related sdist break.

Root causes

• sdist: pyproject [project].readme = "../README.md" — maturin cannot put
  a .. path in an sdist. Now python/README.md.
• wheels: the single ubuntu-latest+macos-latest matrix cross-built /
  emulated everything — aarch64 Linux ran ~6h under QEMU and was killed, and
  x86_64-on-arm64-macOS needed a rustup target that wasn't installed.

Fix

• One runner per native platform: ubuntu-latest (x86_64), ubuntu-24.04-arm
  (aarch64), macos-latest (arm64 native + x86_64 cross). No QEMU.
• CIBW_ARCHS_LINUX=auto64 drops the i686 build (its manylinux image can't
  install rustup — rustup-init needs libatomic.so.1); Rust is installed
  in-container for the Linux manylinux builds.
• MACOSX_DEPLOYMENT_TARGET=11.0 so the x86_64 mac wheel tag matches rustc's
  10.12 floor (delocate had rejected the 10.9-tagged wheel).

Test Plan

A workflow_dispatch run on this branch is fully green in ~5 min —
cargo-test-gate, sdist, and all four wheels (Ubuntu x86_64 + aarch64,
macOS arm64 + x86_64) build and upload artifacts:
https://github.com/Xof/chisel/actions/runs/27104116260

[github-actions]

🚦 Bench results: PR vs main

⚠️ 2 regression(s) detected across 2 scenario/mode pair(s)

Scenario	Mode	Δ throughput	Worst Δ
ycsb-a	chisel-mem	-2.7%	p50 +21.8% ⚠️
mutation-log	sqlite-strict	-2.3%	p99 +10.6% ⚠️
document-store	chisel-mem	+0.2%	—
document-store	chisel-strict	+12.6%	—
document-store	redb-strict	+0.2%	—
document-store	sqlite-strict	-0.2%	—
mutation-log	chisel-mem	-0.9%	—
mutation-log	chisel-strict	+10.8%	—
mutation-log	redb-strict	+0.0%	—
ycsb-a	chisel-strict	+13.9%	—
ycsb-a	redb-strict	-0.0%	—
ycsb-a	sqlite-strict	-0.1%	—
ycsb-b	chisel-mem	+0.3%	—
ycsb-b	chisel-strict	+14.1%	—
ycsb-b	redb-strict	-0.5%	—
ycsb-b	sqlite-strict	+0.5%	—

Per-scenario detail (4 metrics × cells)

document-store

Mode	Throughput	p50	p95	p99
chisel-mem	20849 ops/s → 20900 ops/s (+0.2%)	13.1 µs → 12.9 µs (-1.2%)	135.3 µs → 134.7 µs (-0.4%)	594.2 µs → 586.2 µs (-1.3%)
chisel-strict	2850 ops/s → 3207 ops/s (+12.6%)	22.3 µs → 21.1 µs (-5.6%)	1.09 ms → 972.2 µs (-10.8%)	2.33 ms → 2.02 ms (-13.4%)
redb-strict	4139 ops/s → 4146 ops/s (+0.2%)	19.2 µs → 18.6 µs (-3.1%)	584.5 µs → 580.4 µs (-0.7%)	1.62 ms → 1.56 ms (-3.1%)
sqlite-strict	5111 ops/s → 5101 ops/s (-0.2%)	21.8 µs → 21.7 µs (-0.5%)	439.9 µs → 447.2 µs (+1.7%)	1.39 ms → 1.39 ms (+0.2%)

mutation-log

Mode	Throughput	p50	p95	p99
chisel-mem	40569 ops/s → 40189 ops/s (-0.9%)	27.3 µs → 27.4 µs (+0.2%)	34.3 µs → 34.3 µs (+0.0%)	45.7 µs → 45.1 µs (-1.2%)
chisel-strict	1345 ops/s → 1491 ops/s (+10.8%)	868.1 µs → 806.4 µs (-7.1%)	1.23 ms → 1.12 ms (-8.7%)	2.03 ms → 1.81 ms (-10.9%)
redb-strict	1806 ops/s → 1807 ops/s (+0.0%)	193.8 µs → 198.0 µs (+2.2%)	295.2 µs → 322.3 µs (+9.2%)	30.30 ms → 29.51 ms (-2.6%)
sqlite-strict	4501 ops/s → 4397 ops/s (-2.3%)	133.2 µs → 138.9 µs (+4.3%)	360.0 µs → 388.6 µs (+7.9%)	516.7 µs → 571.7 µs (+10.6%) ⚠️

ycsb-a

Mode	Throughput	p50	p95	p99
chisel-mem	27257 ops/s → 26523 ops/s (-2.7%)	39.6 µs → 48.2 µs (+21.8%) ⚠️	90.8 µs → 91.2 µs (+0.5%)	103.9 µs → 104.7 µs (+0.8%)
chisel-strict	1836 ops/s → 2092 ops/s (+13.9%)	564.0 µs → 517.7 µs (-8.2%)	1.30 ms → 1.15 ms (-11.9%)	1.73 ms → 1.45 ms (-16.3%)
redb-strict	2681 ops/s → 2680 ops/s (-0.0%)	160.2 µs → 159.0 µs (-0.7%)	262.5 µs → 255.4 µs (-2.7%)	465.0 µs → 454.8 µs (-2.2%)
sqlite-strict	145856 ops/s → 145774 ops/s (-0.1%)	7.0 µs → 7.0 µs (-0.4%)	8.9 µs → 8.9 µs (+0.2%)	11.0 µs → 10.8 µs (-2.4%)

ycsb-b

Mode	Throughput	p50	p95	p99
chisel-mem	160739 ops/s → 161262 ops/s (+0.3%)	4.2 µs → 4.2 µs (+1.0%)	47.9 µs → 48.2 µs (+0.7%)	54.6 µs → 54.4 µs (-0.4%)
chisel-strict	18014 ops/s → 20557 ops/s (+14.1%)	6.5 µs → 6.2 µs (-4.5%)	555.4 µs → 518.1 µs (-6.7%)	1.16 ms → 974.3 µs (-15.8%)
redb-strict	27220 ops/s → 27074 ops/s (-0.5%)	2.8 µs → 2.8 µs (+1.1%)	155.2 µs → 155.1 µs (-0.1%)	246.6 µs → 235.8 µs (-4.4%)
sqlite-strict	177272 ops/s → 178181 ops/s (+0.5%)	5.6 µs → 5.6 µs (-0.2%)	7.7 µs → 7.6 µs (-0.4%)	10.0 µs → 10.0 µs (-0.7%)

_{Generated by chisel-bench-diff at 2026-06-07T21:02:26Z. Compares PR HEAD against main. Never blocks merge — signal, not gate. Thresholds: throughput 5%, p50 5%, p95 10%, p99 10%.}