perf: use static strings for static error messages and avoid needless clones

[DXist]

This PR made perf-related changes:

• switches to static string slices and Cow<'static, str> for static error messages
• avoids clones of intermediate results in extract_signatures_inner helper
• avoids intermediate alg clone when only reference is used downstream
• uses owned field_values to avoid intermediate to_vec clone
• takes signatures from the dictionary via swap_remove instead of get + clone
• SignatureBase::try_new avoids a clone of component_lines.
• changes output type of as_field_value to Cow<'static, str> to avoid intermediate allocation for HttpMessageComponentValueInner::String variant.

Stylistic change: as_field_value is renamed to to_field_value because conversion is potentially not cheap due to allocation for KeyValue case.

To measure the effect of the clone-related changes, I also added noncrypto.rs benches module with two benches related to

• signer side SignatureBase construction and
• verifier side header parsing/SignatureBase construction.

I ran the benches before the perf changes and saved the results as my baseline.

Then I ran the benches after the perf changes and got ~8% improvement.

cargo bench -p httpsig --bench noncrypto --  --baseline=baseline
   Compiling httpsig v0.0.24 (/Users/rinatshigapov/workspace/httpsig-rs/httpsig)
    Finished `bench` profile [optimized] target(s) in 17.12s
     Running benches/noncrypto.rs (target/release/deps/noncrypto-33899b8172d49ef8)
Gnuplot not found, using plotters backend
sig base try_new        time:   [2.1496 µs 2.1759 µs 2.2058 µs]
                        change: [−9.1410% −8.5174% −7.8754%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 4 outliers among 100 measurements (4.00%)
  4 (4.00%) high severe
sig base try_parse      time:   [2.5472 µs 2.5860 µs 2.6457 µs]

change: [−8.4396% −7.5125% −6.3287%] (p = 0.00 < 0.05)

Performance has improved.

Found 6 outliers among 100 measurements (6.00%)

3 (3.00%) high mild

3 (3.00%) high severe

It's also possible to avoid cloning of signature_params: &HttpSignatureParam in HttpSignatureBase::try_new and accept owned params in set_message_signatures. If the signer uses a template value for HttpSignatureParam, the cloning cost could be decreased by using compact_str::CompactString.

I tried locally to accept the owned signature_params and swap_remove them from the parsed HeadersMap.

The improvement over the baseline became ~14-15%.

cargo bench -p httpsig --bench noncrypto --  --baseline=baseline
   Compiling httpsig v0.0.24 (/Users/rinatshigapov/workspace/httpsig-rs/httpsig)
    Finished `bench` profile [optimized] target(s) in 17.10s
     Running benches/noncrypto.rs (target/release/deps/noncrypto-33899b8172d49ef8)
Gnuplot not found, using plotters backend
sig base try_new        time:   [1.9997 µs 2.0062 µs 2.0133 µs]
                        change: [−15.587% −15.071% −14.630%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 4 outliers among 100 measurements (4.00%)
  3 (3.00%) high mild
  1 (1.00%) high severe
sig base try_parse      time:   [2.3689 µs 2.3805 µs 2.3924 µs]

change: [−14.595% −13.957% −13.337%] (p = 0.00 < 0.05)

Performance has improved.

Found 6 outliers among 100 measurements (6.00%)

3 (3.00%) high mild

3 (3.00%) high severe

[DXist]

@junkurihara , I rewrote the interface to pass owned signature_params to avoid clones, got ~12% improvement over the baseline code:

cargo bench -p httpsig --bench noncrypto --  --baseline=baseline
   Compiling httpsig v0.0.24 (/Users/rinatshigapov/workspace/httpsig-rs/httpsig)
    Finished `bench` profile [optimized] target(s) in 15.36s
     Running benches/noncrypto.rs (target/release/deps/noncrypto-33899b8172d49ef8)
Gnuplot not found, using plotters backend
sig base try_new        time:   [2.0824 µs 2.0887 µs 2.0949 µs]
                        change: [−12.797% −12.276% −11.820%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 6 outliers among 100 measurements (6.00%)
  6 (6.00%) high mild

sig base try_parse      time:   [2.3976 µs 2.4088 µs 2.4205 µs]
                        change: [−13.093% −12.309% −11.449%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  2 (2.00%) high mild
  1 (1.00%) high severe

[DXist]

I removed intermediate allocations in fmt::Display implementations and reserved a typical capacity for signature base and header value buffers - it gave ~0.5% more improvement.