fix(ui): restore log search option selection highlight

[sjuxax]

Fixes #3004 .

Option rows used theme.text(focused, false), which only dimmed non-focused lines and never applied selection_bg.

Use theme.text(true, selected) for option focus, refresh textarea style/cursor on enabled(), and keep jump-to-SHA field focused (invalid SHA still uses block danger styling).

This code was written by Grok 4.5.

Followed the checklist:

• I added unittests
• I ran make check without errors
• I tested the overall application
• I added an appropriate item to the changelog

[CAOShurong]

Code-grounded verification against the regression mechanism (couldn't build locally on Windows — openssl-sys wants a source build with perl/VC; CI covers the build).

I traced the regression independently of the issue text: since the tui-textarea migration (b9a2e13, first shipped in v0.25.0), TextInputComponent's TextArea is built once in show_inner_textarea() with the style captured at build time, and enabled() only flipped self.selected without touching the cached widget — so the input kept its "active" style and block cursor forever, and the option rows only ever had a default-vs-DarkGray foreground difference (theme.text(x, false) never hits the selection_bg branch). That matches the reported "no visible highlight while moving through options".

This PR addresses both halves of that mechanism:

1. enabled() now calls apply_enabled_style(), which mutates the cached textarea in place (set_style / set_placeholder_style / set_cursor_style REVERSED-vs-none), and show_inner_textarea() applies it after (re)builds — the focus state is finally visible on the input itself, not just the flag.
2. option_line() uses theme.text(true, selected), so the focused option gets the real selection_bg instead of relying on the fg-only cue. Nice side effect: the seven duplicated blocks collapse into one helper.

The new unit test asserting cursor_style REVERSED toggling covers the core of (1).

One deliberate behavior change worth a maintainer's explicit eye (not a defect): in the JumpCommitSha path, find_text.enabled(false) becomes enabled(true) with the comment "Field stays focused; invalid SHA is shown via block style" — the field now stays visually focused in that mode where it was dimmed before. If that's the intended UX for SHA jump mode, this is fine; just making sure it's a conscious choice rather than fallout from the refactor.

Test result on the head: not run locally (build blocker above); CI on this PR is the runtime evidence.