fix: preserve className styles on multi-config and non-style array targets

[Inv1x]

Summary

Fixes two related bugs where className-computed styles were silently destroyed:

• Non-style array targets (e.g. ["contentContainerStyle"]): deepMergeConfig did not merge className + inline styles for length-1 array targets. Inline values simply overwrote the className-computed value instead of producing a style array (the behavior already implemented for the ["style"] path).
• Multi-config components (e.g. ScrollView with both className→style and contentContainerClassName→contentContainerStyle): getStyledProps iterates over multiple configs, and each iteration produced a full props object via Object.assign. Later iterations overwrote earlier iterations' correctly-merged target props.

Reproduction

// Bug 1: contentContainerClassName styles destroyed by inline contentContainerStyle
<ScrollView
  contentContainerClassName="bg-green"
  contentContainerStyle={{ padding: 10 }}
/>
// Expected: contentContainerStyle = [{ backgroundColor: "#008000" }, { padding: 10 }]
// Actual:   contentContainerStyle = { padding: 10 }  ← className styles lost

// Bug 2: className styles destroyed on multi-config ScrollView
<ScrollView className="bg-red" style={{ paddingTop: 10 }} />
// Expected: style = [{ backgroundColor: "#f00" }, { paddingTop: 10 }]
// Actual:   style = { paddingTop: 10 }  ← className styles lost

Fix

1. Add a finalKey merge block for length-1 array targets in deepMergeConfig, mirroring the existing string-target merge logic.
2. Save each config iteration's computed target value in getStyledProps and restore them after the loop completes.

Test plan

• Added tests for ScrollView with contentContainerClassName + contentContainerStyle (non-overlapping props produce array, undefined/null preserves className)
• Added tests for multi-config ScrollView: className + style preserved when contentContainerClassName also present
• Added tests for FlatList contentContainerClassName + contentContainerStyle
• Added test for custom styled() with string target and style={undefined}
• Added test for consumed className sources cleanup
• All 1040 tests pass, typecheck and lint clean

[Copilot]

Pull request overview

Fixes cases where className-derived native styles were being lost when merging with inline props, especially for non-style array targets and multi-config components.

Changes:

• Update getStyledProps to preserve previously-merged target props across multiple config iterations.
• Update deepMergeConfig to merge className + inline styles for length-1 array targets (e.g. ["contentContainerStyle"]).
• Add/adjust tests covering ScrollView/FlatList and multi-config scenarios.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/native/styles/index.ts	Adjusts merge logic for multi-config iteration and length-1 array targets to avoid losing computed styles.
src/tests/native/className-with-style.test.tsx	Adds regression tests for non-style targets and multi-config components to ensure computed styles are preserved.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

[danstepanov]

Thanks for the PR @Inv1x, can you please rebase this given the recent merge of #312

[Inv1x]

@danstepanov done

[danstepanov]

Preemptive review to help triage for the 5.0 release, @marklawlor makes the final call.

Both bugs check out on current main. getStyledProps rebuilds the result per config, so on multi config components like ScrollView the second config clobbers the first one's merged style, and length one targets like contentContainerStyle never merged className with inline styles at all. Branch is freshly rebased and the full suite passes on the merge with main, 1049 tests. One non blocking nit, the new finalKey block also reruns for the plain style target since the guard is only length === 1, worth excluding style in a follow up. Inline style silently destroying className styles on ScrollView is a v4 to v5 regression, so I'd treat this as a blocker.