refactor(store): share the codec line reader and drop the test-only settings clear

[ewanc26]

Summary

Closes the store part of #53. The session and settings codecs each had a copy of the value-line check and of the read-a-line loop; both now use src/store/codec_text.h. The settings store's clear was a wrapper around indigo_file_remove that only the tests called, so it is gone and the tests call indigo_file_remove directly.

What changed

• New src/store/codec_text.h: indigo_codec_has_break and indigo_codec_next_line, inline.
• session_codec.c and settings_codec.c use them. The settings decoder keeps its CRLF tolerance.
• indigo_settings_store_clear removed (it was test-only). indigo_session_store_clear stays, since session_auth.c uses it.

Scope and what was not done

• The small-file helper the issue asks for (read, atomic write, set-aside, remove) already exists in src/store/file.c on main, and the three stores use it. Nothing to move there.
• On current main, jscpd at the issue's --min-tokens 80 --min-lines 12 finds no clones in src/store. The issue's store and session-store clone numbers no longer match main.
• The src/atproto/session_read.c, session_people.c and session_lists.c job functions still have repeated shapes. They are compiled only for 3DS, so this PR does not touch them, and session_auth.c and signin.c are untouched (PR feat(signin): password sign-in finds the PDS from the handle #72).
• Not in this PR: indigo_media_known and indigo_layout_hit from the issue's cppcheck list.

Verification

• make test: 4037 checks, 0 failures, before and after.
• make warnings: clean.
• 3DS cross-build (make with devkitARM, Wolfram v0.38.0): builds.
• Not run on a 3DS or in Azahar.

Notes

Net line count is roughly neutral: 29 lines added and 48 removed in tracked files, plus the new 36-line header.

🤖 Generated with Claude Code

[ewanc26]

Superseded by the replacement PR: same refactor on a fresh branch from main, because this one conflicted with the v0.39.0 pin.