Define the canonical formatter contract

[konojunya]

Background

Task 5 requires one implementation-independent canonical representation for .stack source before the lossless compiler model and formatter are implemented. The previous language text left ordering, comments, escapes, and byte-level output partly advisory.

Implementation

• add a normative formatter specification covering indentation, braces, blank lines, authored declaration/property order, lists, string escapes, comments, UTF-8/LF output, and the final newline
• add a CRLF formatter input plus canonical expected source and normalized IR covering leading, trailing, between-declaration, and nested comments
• define exact-byte, idempotence, and semantic-preservation checks for formatter runners
• validate formatter fixture layout, expected UTF-8/LF bytes, the final newline, and normalized IR in CI

Testing

• python3 scripts/validate-compiler-diagnostics.py
• python3 scripts/validate-formatter-fixtures.py
• check-jsonschema --check-metaschema schemas/*.json
• check-jsonschema --schemafile schemas/normalized-ir.schema.json conformance/valid/*/expected.ir.json conformance/formatter/*/expected.ir.json
• diagnostic expectation schema validation
• existing stack-compiler conformance runner against temporary valid cases for both formatter input and expected source
• git diff --check

Review focus / concerns

• declaration, property, layout-statement, and list order is intentionally preserved rather than sorted so normalized IR order and comment token gaps remain stable
• canonical strings emit direct UTF-8 scalars except escaped quotes and backslashes; \uXXXX spellings are normalized away
• the CRLF input is marked binary only to preserve its fixture bytes through Git; expected canonical output remains reviewable UTF-8/LF text
• compiler-error input is explicitly outside successful formatting, while warnings remain formattable