[rlsw] Fix alpha tracking for persistent current colors

[iztanos]

Fixes #5961.

This follows the root-cause analysis from @HurricanVD in #5961.

The issue suggested preserving hasColorAlpha from the persistent current color
at primitive boundaries. This PR takes a slightly narrower approach: it updates
hasColorAlpha when a vertex copies the current color.

That keeps the flag tied to the primitive contents themselves. A primitive is
marked as containing alpha only if one of its submitted vertices actually has
alpha below 1.0f, while persistent translucent colors still remain correctly
tracked across multiple primitives.

[raysan5]

@Bigfoot71 please, could you takee a look to this PR?

[Bigfoot71]

I propose that we also remove RLSW.primitive.hasColorAlpha |= (color[3] < 1.0f); from sw_immediate_set_color(), and doing this check only in sw_immediate_push_vertex()

[Bigfoot71]

And while we're at it, we could also make the intent explicit, for example:

    // Track whether any vertex in this primitive has alpha < 1.0
    RLSW.primitive.hasColorAlpha |= (vertex->color[3] < 1.0f);

[Bigfoot71]

Yes, there is indeed a logic issue, and the proposed fix is correct.

I just suggested removing a line that becomes unnecessary with this change, and perhaps adding a comment to make the intent explicit for everyone.

Otherwise, looks good to me.

[raysan5]

@iztanos please, could you implement and test proposed changes? thanks!

[iztanos]

Updated, thanks for the review @Bigfoot71.

I removed the old hasColorAlpha update from sw_immediate_set_color() and kept the check only when a vertex is actually pushed. That makes the flag describe the primitive being built, instead of the current color state.

I also added the small comment you suggested near the new check.

Rebased on current master and reran the small rlsw.h regression test in Docker. The persistent translucent color case still renders both triangles blended.