Cut 0.10.0: consolidate the changelog and refresh every published number

[kurok]

Cuts 0.10.0. Version 0.9.0 → 0.10.0: no API changed, so not a major; one documented behaviour did — lazy modes now pin the payload they borrow from (#239) — which is more than a patch. Single-sourced from Cargo.toml; both lockfiles follow.

The changelog needed consolidating before it could be released

Twenty-two PRs had each appended their own section heading, so [Unreleased] held three ### Added blocks and three ### Changed blocks, and #223's entry appeared twice, verbatim. It is now one Added, one Changed and one Fixed — 21 entries — with the lazy-borrow behaviour change leading Changed, because it is the only one a caller can observe.

Every published number was roughly half of what the code does

All the Readme's performance figures dated from the 0.9.0 measurement. Stale numbers that are 2× wrong are worse than no numbers, so all three tables are re-measured from a single dispatched benchmark run on one runner (EPYC 7763, 4 vCPU), each table keeping its own statistic — minimum for the library comparison, median for the mode and batch tables.

	0.9.0	0.10.0
against other libraries	0.59 ms · 25x / 34x	0.34 ms · 42x / 62x
parse_email	0.575 ms	0.326 ms
parse_email_tree	0.556 ms	0.326 ms
8 × 767 KiB, threads=1	4.48 ms	2.63 ms
2000 × 0.8 KB vs a thread pool	11.7x	13.9x

The mail-parser row also moves 4.6.4 → 4.6.5, which #263 bumped.

One ratio moved the wrong way, and the text now says so. The mode= multipliers shrank — lazy 7.2x → 5.8x, metadata 11.3x → 6.4x — because the default mode roughly halved, so there is less of it left to skip. Every mode is faster in absolute terms than it was in 0.9.0, and the Readme says that rather than quietly dropping the comparison.

Verification

• 867 Python tests including the Readme and docs snippet tests, which execute the examples
• built wheel reports 0.10.0; check_bench_lockfile.py passes (it caught bench/Cargo.lock still on 0.9.0 mid-change)
• Cargo.lock and bench/Cargo.lock regenerated

Not in this PR

The Rust toolchain pin is still 1.98.0; 1.98.1 is out and toolchain-ab.yml is measuring it now, per the procedure in rust-toolchain.toml. That is a separate change with its own measurement — this repo has seen a rustc minor move performance 15–96% (#120) — so it should not ride along with a release. If it measures clean and you want 0.10.0 built on it, merge that first and I will re-tag.