error: 3-4% faster builds via macro instead of trivial '?'

[exrok]

I'm not sure if for code quality reasons, we actually want to do this for the relatively small gain but
thought I put it up just in case, feel free to close.

I've only done trivial try transforms in this PR, but I think there is more savings that could be had with a similar macro to avoid many of the map_err( instances.

This commit introduces the rtry! macro and uses it throughout the codebase instead of ? for forwarding result types whose error types are already same. This is a trick that serde uses: https://github.com/serde-rs/serde/blob/fa7da4a93567ed347ad0735c28e439fca688ef26/serde/src/lib.rs#L236-L248

The de-sugared try operator ('?') invokes the Try and From trait but for most of the error forwarding sites we need neither.

For rustc 1.94.0 (4a4ef493e 2026-03-02):

The total debug build cargo llvm-lines of jiff, reduced from 174096 to 168315 LLVM lines.

Then measured from 5 samples of perf stat on a direct rustc invocation to build the jiff rlib (non incremental), shows the following:

Before (using the try operator)

    10,889,052,814      cycles:u        ( +-  0.08% )
    15,171,370,701      instructions:u  ( +-  0.01% )
     2,671,428,910      branch:u        ( +-  0.01% )

    1.60105 +- 0.00152 seconds time elapsed  ( +-  0.09% )

After (using the macro)

    10,454,390,370      cycles:u        ( +-  0.12% )
    14,591,413,777      instructions:u  ( +-  0.01% )
     2,567,871,039      branch:u        ( +-  0.01% )

    1.55956 +- 0.00335 seconds time elapsed  ( +-  0.21% )

The above perf results show an improvement of 3-4% percent.

System Specs if Needed

rustc:   rustc 1.94.0 (4a4ef493e 2026-03-02) (LLVM 21.1.8)
os:      Arch Linux (kernel 6.18.13-arch1-1)
cpu:     AMD Ryzen 9 5950X 16-Core Processor (32 threads)
memory:  63 GB
caches:  L1d: 512 KiB (16 instances), L1i: 512 KiB (16 instances), L2: 8 MiB (16 instances), L3: 64 MiB (2 instances)

Exact perf stat rustc command I used

env CARGO_PKG_HOMEPAGE='' CARGO_PKG_LICENSE='Unlicense OR MIT' CARGO_PKG_LICENSE_FILE='' CARGO_PKG_NAME=jiff CARGO_PKG_README=README.md CARGO_PKG_REPOSITORY='https://github.com/BurntSushi/jiff' CARGO_PKG_RUST_VERSION=1.70 CARGO_PKG_VERSION=0.2.23 CARGO_PKG_VERSION_MAJOR=0 CARGO_PKG_VERSION_MINOR=2 CARGO_PKG_VERSION_PATCH=23 CARGO_PKG_VERSION_PRE='' CARGO_PRIMARY_PACKAGE=1 LD_LIBRARY_PATH='/tmp/rust-target/debug/deps:/home/user/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib' perf stat -r 5 -e cycles,instructions,branch /home/user/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/bin/rustc --crate-name jiff --edition=2021 src/lib.rs --error-format=json --json=diagnostic-rendered-ansi,artifacts,future-incompat --diagnostic-width=191 --crate-type lib --emit=dep-info,metadata,link -C embed-bitcode=no -C debuginfo=2 --allow=unexpected_cfgs --check-cfg 'cfg(docsrs_jiff)' --cfg 'feature="alloc"' --cfg 'feature="default"' --cfg 'feature="perf-inline"' --cfg 'feature="std"' --cfg 'feature="tz-fat"' --cfg 'feature="tz-system"' --cfg 'feature="tzdb-bundle-platform"' --cfg 'feature="tzdb-concatenated"' --cfg 'feature="tzdb-zoneinfo"' --check-cfg 'cfg(docsrs,test)' --check-cfg 'cfg(feature, values("alloc", "default", "js", "logging", "perf-inline", "serde", "static", "static-tz", "std", "tz-fat", "tz-system", "tzdb-bundle-always", "tzdb-bundle-platform", "tzdb-concatenated", "tzdb-zoneinfo"))' -C metadata=6b7754662bf2080e -C extra-filename=-c5cdb0b61b123153 --out-dir /tmp/rust-target/debug/deps -C linker=/usr/bin/clang -L dependency=/tmp/rust-target/debug/deps -Clink-arg=--ld-path=/bin/wild

Replacement Process:
Initially I tried to use debug info to find trivial From invocations of error types, on the same line as a try operators to constrain the replacements but I ended up just doing aggressive replacements using tree-sitter then reverting the cases that failed...

As you can see from the initial CI failures and patch commits the process has been a bit iterative. I am trusting that CI type checks all code paths, I don't know if I should be. If the premise of this PR is something that could be accepted, I could spend the time to auditing each replacement, just say the word.