doc: note the Windows CRLF/BerkeleyDB build failure in CLAUDE.md

[PrestackI]

CLAUDE.md documents only the Linux-native build. On Windows that path fails in a way whose cause is not obvious, and nothing in the tracked docs warns about it.

The failure

src/bdb53/dist/configure is a POSIX shell script. .gitattributes marks the tree * text=auto with no eol=lf override for src/bdb53, so a clone made with Git-for-Windows' default core.autocrlf=true checks that script out with CRLF. The bundled BerkeleyDB build then dies on it.

Two details make it easy to misdiagnose, and the wording here is deliberate about both:

• It is not a configure-time failure. BerkeleyDB is wired through ExternalProject_Add with a CONFIGURE_COMMAND (src/bdb53/CMakeLists.txt:67-70), and an ExternalProject's configure step runs at build time. cmake -B build returns 0; cmake --build build is what fails. A contributor whose configure succeeded would otherwise conclude the note does not apply to them.
• It does not block all compilation. leveldb and secp256k1 are plain add_subdirectory targets with no dependency on BerkeleyDB_Project, so under -j they compile concurrently with the failure.

The change

Adds a short ### Windows Contributors subsection after the Linux-native build block, pointing at a WSL-native clone as the fix.

It also scopes the two existing Windows references, which currently mislead:

• doc/build.md ("Linux Native") is named as the reference for the normal development flow.
• doc/build-windows-wsl.md is scoped to the win64 case only. Its sole build command is ./build_targets.sh TARGET=win64 and it verifies build_win64/src/gridcoinresearchd.exe — it produces no build/ tree, so it does not serve the ctest --test-dir build loop that CLAUDE.md documents directly below.
• doc/build-msys2.md is deliberately not linked: it is a three-line deprecation stub that redirects to the WSL guide, and MSYS2 builds against a Windows-filesystem checkout, i.e. it reproduces the very failure being described.

The gridcoin-gui-wsl skill is mentioned for Qt GUI iteration, qualified as GUI-only since it configures -DENABLE_TESTS=OFF.

Scope

Markdown only — 17 added lines in one file, nothing removed. No source, CMake, test, workflow, or lint configuration is touched.

Not verified

The quoted error string Syntax error: newline unexpected was not reproduced end-to-end. The CRLF cause is confirmed directly (od -c on the checked-out file shows #!/bin/sh\r\n), but the exact message is a dash/ash rendering and may differ under another shell. Happy to soften it to "a shell syntax error" if a maintainer prefers not to assert a specific string.

If src/bdb53/** eol=lf is ever added to .gitattributes, this subsection becomes obsolete and should be deleted rather than updated.