Speed up rendering and close a few compile-time holes

[gregjotau]

Why

Render speed is now measured. The first pass over HtmlOutput was still
per-character with a flush check on every byte, integers walked a divisor,
and application-relative TrustedUrls allocated even when the context path
was empty. Spring also flushed a 1KiB buffer. A few dead-code and
reference checks were inconsistent with the rest of the compiler.

What

• Single-pass HtmlOutput.text and reverse-digit text(long)
• RenderContext.resolveUrl returns the same string when the context path is empty
• Golden HTML plus a reference encoder, including tiny-buffer equality
• failOnUnusedFragments also fails unused fragment parameters
• #id targets on href / hx-target / hx-include / hx-indicator must exist
• Spring HtmlOutput buffer matches the 16KiB runtime default

Local JMH: inbox ~1.55µs → ~1.02µs, catalog/50 ~5.4µs → ~3.7µs, integers ~1.0µs → ~0.43µs.

How to check

./gradlew test
./gradlew :benchmark:jmh

[gregjotau]

Closing as a stale, conflicted bundle rather than merging it wholesale. The safe unused-fragment-parameter part shipped in #46. The proposed static hx-target/hx-include/hx-indicator ID validation was tested against all consumers and produced false positives for conditional targets, so it needs reachability-aware DOM modeling before it can be sound. The rendering performance changes should return as a focused PR with benchmark evidence.