Consider making LF the default line terminator

[matt-edmondson]

The decision that was deferred

#81 made the line terminator configurable but kept the default as Environment.NewLine, so determinism is opt-in. That was a deliberate deviation from its own first acceptance criterion, which asked for byte-identical output across platforms by default.

The reasoning at the time: flipping the default silently changes the bytes every current Windows consumer gets, and that is a maintainer's call rather than an implementer's. This issue is that call, written down so it does not just get lost.

The case for flipping it

• It is a code-generation library. Its output is overwhelmingly written to files, committed, diffed, or compared against golden files — all of which want reproducibility more than they want the host's convention.
• The status quo produces silent bugs. ktsu.Semantics carried a post-processing pass for years purely to undo Environment.NewLine, and — see Retire GeneratedSource once CodeBlocker emits deterministic newlines Semantics#184 — that pass was itself rewriting to the wrong terminator after the repository migrated to LF, with git quietly normalising the difference on every commit. Nobody noticed because nothing surfaced it.
• The tests here made the same mistake. 46 of 67 failed on any non-Windows host, and only ever passed because CI ran on Windows. That was fixed independently on main in d4a7153.
• LF is what the ecosystem stores. This repository's own .gitattributes says * text=auto eol=lf.

The case against

• It is a behavioural break for any consumer relying on the current output, in a way that will not show up as a compile error — only as a diff.
• Consumers who genuinely want the host convention would have to opt in, whereas today they get it for free.

If it goes ahead

• One-line change in CodeBlocker.cs: NewLines.Host → NewLines.Lf in the three defaulting call sites.
• Tag the commit [major], since the release pipeline derives the version from it.
• README.md needs its "Line Endings" section reworded — NewLines.Host becomes the opt-in rather than the default.
• The existing tests are already terminator-explicit, so they should not need changing. NewLineTests.DefaultNewLineStringIsTheHostTerminator is the one that pins current behaviour and would be rewritten.

Acceptance criteria

• Decision recorded here either way, so it stops being an open question.
• If flipped: default is LF, [major], README.md updated, and the change called out in release notes as a behavioural break.

Context

Deferred from #81, delivered in #87.