perf: scan multi-prefix token detectors once

[tboser]

Summary

• scan each same-kind prefix family in one pass instead of rescanning the full input once per prefix
• skip prefix comparisons cheaply when the current byte cannot begin any configured prefix
• cover every GitHub, Slack, Stripe, and OpenAI prefix variant with synthetic fixtures

Why

The four multi-prefix detectors currently account for 20 complete passes over a clean input. This is visible when a caller selects credential detectors without the PII set.

On an Intel Xeon 8488C, with a synthetic clean 8 MiB JSON body and the same 14-detector credential-only set before/after this commit:

Case	Before	After
full set, serial, one caller	17.6 MiB/s	93.4 MiB/s
full set, parallel, one caller	70.8 MiB/s	316.5 MiB/s
full set, parallel, four callers	172.4 MiB/s	552.2 MiB/s
GitHub detector	64.9 MiB/s	2,275.9 MiB/s
Slack detector	65.0 MiB/s	2,326.8 MiB/s
Stripe detector	63.6 MiB/s	2,237.4 MiB/s
OpenAI detector	189.7 MiB/s	2,339.3 MiB/s

Inputs and credentials used for tests and measurement are synthetic.

Validation

• cargo fmt --all -- --check
• cargo clippy --all-targets -- -D warnings
• cargo test --all-features
• cargo build --no-default-features
• cargo test --release --test complexity -- --ignored

[ptukovar]

Thank you 😸