Fix/open page by slug missing md projection

[SSamDav]

What this PR does

When a new device joins a sync chain, the peer ops are replayed into the in-memory CRDT tree by reload_workspace, making every synced page visible in search — but only today's journal gets its .md file projected to disk. open_page_by_slug then calls build_page_view → read_page_outline, which does fs::read_to_string().unwrap_or_default(): a missing file silently returns "", parse("") produces an empty outline, and the page opens blank with no error. The fix calls apply_page_md_with_sidecar before building the view (mirroring what open_ref already does), so the .md + sidecar are written to disk on first open if absent.

How to verify

cargo test --workspace --all-targets
cargo clippy --workspace --all-targets -- -D warnings
cargo fmt --all -- --check

Manual smoke (iOS + macOS):

1. On device A, create several pages with content.
2. Pair device B (fresh install) to device A via the Devices sheet.
3. On device B, open the page picker (Cmd+P / search) — pages from device A should appear.
4. Tap any of those pages — they should open with their full content, not an empty body.

Related issues / docs

Closes #120

Anything reviewers should look at carefully

• No CRDT-correctness implication: this change only adds a lazy .md projection before a read; it does not touch op generation, apply_op, or the tree structure.
• No markdown-format implication: apply_page_md_with_sidecar is the same writer used by open_ref and the journal renderer; the output format is unchanged.
• The projection failure is non-fatal on both clients: a warn! / eprintln! is emitted and build_page_view still runs (returning empty content as before, rather than crashing).
• No new public API.

Out of scope for this PR

• Proactively projecting all synced pages to disk during reload_workspace (would fix the same symptom for any navigation path, but could be expensive on large workspaces — separate issue).
• The open_journal_for command has a similar shape but journal pages are always projected at boot by reproject_page, so it is not affected.'

[avelino]

Two changes are riding in one PR. The #120 fix (desktop + mobile) is the right shape and mirrors open_ref cleanly. The TUI slash-palette reorder is the problem: it's the same change as #125, which is still in review for the exact bug it still carries here.

Split it. Keep #126 as the #120 fix only, move the TUI part back to #125. Then #126 merges once the sidecar churn below is handled.

blocking

• The TUI reorder only syncs the empty-query case. With an active query, arrow navigation still walks candidates in score order while the renderer always buckets by category, so ↓ can move the highlight up. Same bug I flagged on #125.

needs fixing

• open_page_by_slug now projects .md + sidecar on every open. build_sidecar stamps last_synced_at: now(), so the .outl rewrites on every page open even when the .md is byte-identical. This is the hottest nav path, so it turns into sync churn. Guard it to only project when the file is missing.

No new test for #120 either. A cross-category navigation test locks the palette regression; a "synced page opens with content" test locks #120.

[avelino]

This projects on every open. build_sidecar stamps last_synced_at: now(), so the .outl gets rewritten every time you open a page, even when the .md is identical. That's pure sync churn on the hottest nav path.

Guard it: only project when pages/<slug>.md is absent (resolve via page_meta + journal::page_md_path). Same fix on the mobile mirror at line 141. Worth an issue to guard open_ref the same way, it has the same unconditional projection and nobody caught the churn there.