test(cli): assert on what the CLI owns, not on what the OS says - #555

Two of the eighteen known windows failures were tests asserting text that was never ours to depend on.
Neither is a product defect: the CLI behaves correctly on windows in both cases.

TestCLI_ConfigPath

cli_test.go:324: output missing "devlore/config.yaml" (len=73)

The printed path is OS-native, so the literal only ever matched on Unix -- windows prints
devlore\config.yaml. Under the identity/access rule a config path is access, so the CLI is right and
the test encoded a Unix assumption.

It now asserts xdg.ConfigPath("devlore", "config.yaml").

Asking xdg rather than cli.SharedConfigPath() is deliberate: SharedConfigPath is the very function the
command calls, so asserting against it would agree with itself however wrong it became. xdg is one level
down and states the claim independently.

TestCLI_RunMissingFile

cli_test.go:132: output missing "no such file" (len=1091)

The runner wraps os.ReadFile with %w:

return nil, fmt.Errorf("reading script %s: %w", r.script, err)

so the tail of that message is the OS speaking -- no such file or directory on Unix, The system cannot find the file specified. on windows. The test was asserting which platform it ran on.

It now asserts the two substrings we produce: the reading script prefix, and the path the caller passed.

Accepted trade: the assertion can no longer distinguish a missing file from an unreadable one. This is a
CLI smoke test, and the wrapped error still carries the OS's specific diagnosis for anyone reading it. The
alternative -- normalizing fs.ErrNotExist into our own phrasing -- buys uniform text at the cost of a
precise one, and works against the %w idiom.

Verified

• make check -- clean
• Windows baseline expected to move 18 -> 16, verified name-for-name against develop before merge

Refs #373.