Add %{field} escape to -printf / -printfln / -fprintf / -fprintfln

[helly25]

-printf's format is a %-directive language and treats a bare {...} literally (printf formats legitimately contain braces, e.g. JSON), so the rich brace field vocabulary was unreachable from these per-branch actions - unlike the whole-run --format global. This bridges the gap with an xff %{NAME} / %{NAME:qualifier} escape.

%{...} renders through the same fields::Template engine as --format braces, so the full vocabulary comes along:

xff . -type f -printf '%{relpath}\t%{size:h}\t%{mtime:%Y-%m-%d}\n'
xff . -type l -printf '%f -> %{target}\n'
xff --define=B=/backup . -type f -printf '%{relpath} vs %{def.B}/%{relpath}\n'

%{relpath}, %{core}, %{suffix}, %{target}, %{def.NAME}, %{env.NAME}, %{capture.N}, time qualifiers, the s/PAT/REPL/ rewrite, path-component qualifiers - all work.

Rules

• %% stays a literal percent; a bare {...} stays literal (find-compat, no breakage); an unterminated %{ is emitted literally, matching the field template's own lenient handling.
• The escape is an xff extension inside otherwise find-native actions, so EnforceStyle rejects a %{...} format under --config=find (a plain %p format stays fine). -printfln / -fprintfln are xff already.
• The symlink read backing %{target} is skipped unless the format actually uses %{.

FormatPrintf now takes the EvalContext (for captures/defines/env/target), like RenderExecArgv.

Self-doc + tests: -printf summary notes the escape; --help / --man / --markdown regenerate from the registry. run_test pins the escape rules (expand / %% / bare-literal / unterminated); parser_test pins the strict-find rejection; a dogfood xff_golden case pins the find(reject)-vs-xff(expand) divergence. Full bazel test //xff/... green (49); both clang-format versions + buildifier + pre-commit clean.