fix(repair-encoding-cli): reconfigure stdio to UTF-8 like the other entry points

[arnoldwender]

What and Why

mempalace/_stdio.py states the rule in its own docstring:

Every console entry point that touches stdio needs to fix this on Windows -- the MCP server, the CLI, the fact_checker --stdin mode -- so the reconfigure code lives here in one place.

cli.py:2160 and fact_checker.py:321 both apply it. scripts/mempalace_repair_encoding.py, added in #2148, does not (grep -c _stdio = 0).

That gap lands hardest on this particular entry point. The tool exists for Windows users whose palace carries legacy mojibake, and _print_change emits a before/after preview for every proposed change. Under the console codepage the tool was written for, that preview cannot be encoded — the lead bytes of the corruption it detects are precisely the characters the codepage rejects — so the run dies before repairing a single drawer.

Reproduction

With stdout on a legacy codepage, printing the same preview the tool emits:

UnicodeEncodeError: 'gbk' codec can't encode character '\xc3'
in position 14: illegal multibyte sequence

U+00C3 is Ã — the lead character of the family this tool repairs.

Expected: the dry-run lists proposed changes and exits 0.
Actual: exits 1 on the first change containing a non-ASCII preview.

Change Summary

• scripts/mempalace_repair_encoding.py — add _reconfigure_stdio_utf8_on_windows() (thin wrapper over the shared helper, mirroring cli.py / fact_checker.py) and call it first thing in main().
• tests/test_repair_encoding_cli.py — new file, loading the script by path the way tests/test_backfill_authored_at.py does.

replace on stdout/stderr, matching the other two console entry points: the preview carries verbatim drawer text that may hold surrogate halves round-tripped from filenames, where strict would crash mid-preview.

Test Plan

• test_reconfigures_stdio_to_utf8_on_windows — mirrors the existing assertions in test_cli.py and test_fact_checker.py, pinning the per-stream errors policy.
• test_change_preview_survives_a_legacy_console_codepage — reproduces the actual failure rather than asserting a call happened: a child process gets a legacy codepage for stdout and is asked to print a real mojibake preview.
• Mutant-checked. Neutralising the reconfigure call turns the second test red with the genuine UnicodeEncodeError: 'gbk' codec can't encode character '\xc3', so it is pinning the behaviour and not the wiring.
• Full suite: 3861 passed, 31 skipped.
• ruff check . and ruff format --check . clean (213 files).

Notes

Related to #1122 (same exception class, main CLI's help output) but not a duplicate: different entry point, and a different fix — that one can be solved with ASCII-safe static strings, whereas this one prints arbitrary user content and needs the stdio reconfigure.

Found while auditing the encoding-repair surface. I also filed #2193 for a separate, more serious defect in the same module — that one is a correctness question in _HIGH_CONFIDENCE_RUN and needs a design call, so I kept it out of this PR to keep this one trivially reviewable.

[igorls]

Wave 1 for 3.7.0: LGTM.