perf: skip builders for single-spec formats

[He-Pin]

Motivation:

PR #776 showed that format-heavy workloads benefit when the format path avoids unnecessary intermediate assembly. This split keeps only the smallest safe idea: short format strings with exactly one specifier and no static literal text (for example %08d, %010x, %-20s, %20s) do not need a StringBuilder after the formatted value has already been computed.

Key Design Decision:

Keep the existing generic formatting implementation and all validation/arity checks. The optimization only bypasses appending to StringBuilder after the single formatted value is known, so format semantics and error behavior stay unchanged.

Modification:

• Detect specBits.length == 1 && parsed.staticChars == 0 in Format.format.
• Avoid allocating/appending a StringBuilder for that case.
• Return the computed formatted value directly after the existing too-many/too-few value checks.

Benchmark Results:

JMH (./mill -j 1 bench.runRegressions ..., ms/op lower is better; ops/ms higher is better):

Case	master ms/op	PR ms/op	master ops/ms	PR ops/ms	Delta
repeat_format	0.190	0.133	5.263	7.519	+42.9%
large_string_template guard	1.155	1.160	0.866	0.862	-0.4% noisy/neutral

Scala Native hyperfine (hyperfine --warmup 10 --min-runs 50 -N, ms lower is better):

Case	master native	PR native	jrsonnet	Result
repeat_format	6.4 ± 0.9 ms	6.4 ± 0.7 ms	5.6 ± 1.0 ms	Native neutral; JMH target positive
large_string_template guard	12.6 ± 7.4 ms	11.4 ± 0.8 ms	5.4 ± 0.9 ms	No Native regression observed

Analysis:

The target case is dominated by many short format expressions. Returning the already computed formatted string removes a redundant builder allocation/append path on the JVM. The guard case does not use this single-spec/no-static-literal path and remains effectively unchanged within benchmark noise.

References:

• Source idea: perf: comprehensive Scala Native render pipeline optimization #776
• Split branch commit: He-Pin/sjsonnet@1f58504d

Result:

• ./mill -j 1 __.reformat && ./mill -j 1 __.test passed locally.
• Draft PR split from the broader perf: comprehensive Scala Native render pipeline optimization #776 optimization branch to keep the change reviewable and avoid carrying unrelated Native template work.