Respect merged PSM1 formatting scope

[PrzemyslawKlys]

DefaultPSM1 no longer formats a root PSM1 produced by module merging or missing-function inlining. That output uses OnMergePSM1; leaving merge formatting disabled now preserves the merged text and avoids an unrequested formatter pass.

Ordinary staged PSM1 files and project-root files retain their default settings. Explicit merge formatting still takes precedence and runs once. PSD1 selection is unchanged.

This was reproduced while validating the Locksmith2 build upgrade: the wrapper enables DefaultPSM1 but leaves OnMergePSM1 disabled. The correction belongs in the shared pipeline, so no consumer workaround or PSScriptAnalyzer update is required. Consumers need a public PSPublishModule package containing this fix to use it.

Validation: the regression case fails before the change; 28 focused tests pass afterward, including scope selection and existing formatter/merge contracts. Five scope cases also verify settings precedence and preservation of skipped bytes. The native module package builds successfully. Independent read-only review found no actionable defects.

Real before/after builds retain Locksmith's exact merged PSM1 bytes. Locksmith2 retains identical parser tokens after excluding comments, line breaks, continuation markers and end-of-input; both modules import with 1/9 public functions on Core and Desktop. The existing Locksmith2 post-build suite passes 32/32.

Local cached Locksmith2 builds measured 20.89 -> 15.06 seconds and 18.06 -> 12.14 seconds median across two fixed CPU cache domains, with one warmup and three rotated samples per lane/domain using unchanged PSScriptAnalyzer 1.24.0. All 12 measured builds pass artifact validation. An earlier incomplete domain run hit disk exhaustion during an isolated post-build copy; that entire run was excluded and both lanes were rerun after cleanup. These workstation results compare the previous and corrected 3.0 selection logic, and do not establish a general 3.0-versus-2.0 ranking.