yes (and no). it uses both, but in different ratios.
There is a single global light source to allow for bump mapping. the actual lighting itself uses this plus the incoming lightmap.
The lightmap is pixel based so there is no actual source - just the light itself.
But...
for every 3d object. there is a color morphed with a texture plus a lightmap plus global lighting plus bump mapping and finally some highlighting.
These are controlled per frame and per object, so all or any input can be changed. in practice this is done in the tile editor, where color/texture/bump is available via a slider for simplicity
Here's a shot with a pillar object

- all 3 pillars have the same base color - they could be different
- the first has no texture and no bump/highlight
- the second has just bump/highlight added at full
- the third has just texture added at full
All of these are mixed to get the required 'look'
Texturing and bump mapping come from a single uber texture. there is only one texture used for EVERYTHING. All models get their textures from this single image. It is currently rgb, but could be just a single channel with the 2 other channels used for other information if required.
Base bump mapping comes from a single normal map for everything with only one texture.
Everything (model data and texture data) is send to the GPU at startup and then kept there.
the only data that is sent to the GPU each frame is what object are wanted, where, and their surface information - several floats per object. NO 3d data or texture data is ever sent (it's already there)