Add replay chart v2 with ceiling/GF overlays and fullscreen (PRO-51)

[shaheinm]

Summary

Phase 4 of the dive replay engine (PRO-23). Ports the animation work from closed PR #160 forward onto latest main now that PRO-60 (deco accuracy) and PRO-61 (actual-dive samples) have reshaped ReplayProfileSheet.

• ReplayChartData precomputes depth, ceiling, GF99, SurfGF, gas-switch, and deco-stop data with O(1) scrub lookups.
• ReplayChartFullscreenView presents the chart landscape with overlay toggle chips (Ceiling, GF99, SurfGF).
• ReplayProfileSheet wires the chart into the existing parameter sheet.
• planned_stops added to ProfileGenResult (Rust + UDL + Swift bindings) to surface the pass-1 stop schedule for chart band visualization.

Closes PRO-51. Supersedes #160 (closed due to ReplayProfileSheet conflicts from PRO-60/61). Completes the "Replay Complete" milestone and unblocks PRO-53 (Phase 5 comparison view).

Tests

• Rust (2 new): test_planned_stops_surface_deco_dive, test_shallow_no_deco (extended) — ordering, duration, and sample-hold correspondence.
• Swift (2 new): FFI round-trip for plannedStops on both shallow NDL (empty) and deep conservative-GF (non-empty) profiles.
• Full suites green locally: Rust 303 pass, Swift 400 pass, SwiftLint 0 violations.

Test plan

• Open a dive, Replay, configure params, Generate, verify stat cards + chart appear
• Play, verify progressive drawing at selected speed (1x–60x)
• Pause, drag slider, verify scrub
• Change speed mid-animation
• Reset clears to time 0
• Fullscreen + landscape, overlay chips (Ceiling, GF99, SurfGF) toggle
• Close fullscreen via X button and via swipe gesture, verify timer pauses
• No-deco dive: no ceiling/stops, clean depth line
• Deep deco dive with gas switches: stop bands + switch markers render correctly
• Shallow dive: empty stop bands

Follow-ups

• PRO-53 (Phase 5): side-by-side comparison of real vs simulated stats

Made with Cursor

[codecov]

Codecov Report

❌ Patch coverage is 94.44444% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
core/src/deco/profile_generator.rs	94.44%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[shaheinm]

Second-opinion review (gpt-5.3-codex) — REQUEST-CHANGES → addressed in d6e6d47

Ran an independent review with a gpt-5.3-codex Cursor subagent instead of the Codex MCP (follow-up commit bac8b6a updates CLAUDE.md step 8 to reflect that multi-model subagents supersede the external MCP).

Findings → fixes

Blocker

• Actual-dive mode ran the second deco engine call (planAscent: true) but the actual-dive ReplayChartData initializer hardcoded decoStopBands = [] — the planned stops were silently dropped before rendering.
  → Fixed: ReplayChartData(samples:decoResult:plannedStops:gasMixes:depthUnit:) now accepts the planned stops and matches them against sample depths (1.0 m tolerance, max-depth sample as bottom-end proxy).

Should fix

• Dual engine calls were serialized inside a single detached task.
  → Fixed: async let + two detached calls; roughly halves latency on longer dives.
• Actual-dive initializer assumed monotonic tSec but didn't enforce it (scrub + band matching use binary search).
  → Fixed: defensive sorted { $0.tSec < $1.tSec } at entry.
• PRODUCT_BUNDLE_IDENTIFIER changed from azlucis.Profundum → com.azlucis.Profundum in project.pbxproj. This was accidental Xcode drift from saving the pbxproj for the new files, out of scope for PRO-51 (would affect signing/provisioning).
  → Fixed: reverted both occurrences.

Nits (not addressed in this PR)

• ReplayChart.swift at ~1140 lines could be split into data vs. rendering. Worth a dedicated refactor ticket; not blocking.
• Explicit VoiceOver hint on the accessible scrub slider — small, can follow up.

Mutation-testing note

• core/mutants.out shows no missed profile_generator mutants but one timeout in generate_ascent (line 492). Worth tracking but not a regression introduced by this PR.

All local checks re-run: Rust 303/303 pass · Swift 400/400 pass · macOS build green · SwiftLint clean.

[shaheinm]

Second-opinion review round 2 (gpt-5.3-codex) — no blockers → addressed in acc7b29

Fresh independent review of the current head after the round-1 fixes. Confirmed all round-1 fixes are correct (planned stops wired in actual mode, parallel engine calls, defensive sort, bundle id intact).

Findings → fixes

Should fix

• Actual-dive scrub readout: resultSummary passed unsorted samples.toSampleInputs() to ReplayChartSection, but nearestTSec binary-searches assuming monotonic tSec — out-of-order imports would get corrupted scrub values.
  → Fixed: sorted before mapping, matching ReplayChartData's defensive sort.
• extractResultData used unsorted samples.last?.tSec for Recorded Time.
  → Fixed: samples.map(\.tSec).max().

Found in self-review

• The empty-params guard in generateFromActualSamples returned silently after generate() set isGenerating = true, leaving the spinner stuck if ever hit.
  → Fixed: resets isGenerating and surfaces an error message.

Nit (not addressed)

• Task.detached inside Task + async let is redundant; plain async let would be cleaner. Cosmetic, leaving as-is.

Local checks: SwiftLint clean, macOS build green (only pre-existing warnings in unrelated files).