feat: show_hints — opt out of the hint chrome

[teocns]

Closes #4.

@scc02 asked for a way to hide the ⟪neocursor · <Tab> jump⟫ label while keeping the suggestion itself. This adds show_hints, using the exact API shape proposed in the issue.

require("neocursor").setup({
  show_hints = false,                  -- hide both hint surfaces
  -- show_hints = { edit = false },    -- hide the label, keep the jump pill
})

Two surfaces, and they aren't equally safe to hide

Surface	Renders	Cost of hiding
edit	⟪neocursor · <Tab> accept⟫	none — the diff beside it already shows the change
prediction	⟪<Tab> → L42⟫	the only on-screen sign a jump is queued

false hides both, since that's the least surprising reading of the option name. The table form keeps the pill for anyone who wants the label gone but still wants to see where <Tab> will land. The README documents the tradeoff rather than hiding it.

The trap

show_prediction() does double duty — it paints the pill and returns the boolean that gates jumping at four call sites (init.lua 358, 398, 814, 956).

The obvious implementation (return false when hints are off) renders correctly and silently disables <Tab> jumping. So: the paint is conditional, the return value is not.

Verification

Two independent angles, both in CI on ubuntu/windows/macos:

• test/hints_spec.lua (new, 16 assertions) — label present/absent, diff body byte-identical either way, prediction-pill namespace isolation, all 7 normalization cases.
• test/flow_spec.lua now reruns its entire 20-assertion behavioral suite under NEOCURSOR_SPEC_NO_HINTS=1. Tab jumps to predicted line and prediction consumed by jump pass with zero chrome — that's the regression that would otherwise ship silently.

:NeocursorDebug prints the resolved setting.

Note

M._normalize_hints is exposed (underscore-prefixed, internal) so the spec can test the pure normalizer without spawning a sidecar.