You are correct -- the layer metadata is not updated by SQL inserts. But this isn't particularly significant, since the layer extent ought to be initialized to the range of valid values before population. And if it isn't, it may be for a reason that changing the index as part of insert would impact, so not changing it is probably safer.
If you do need to update the extent of a layer, there are a number of ways to accomplish this, but the safest use the ArcSDE API (either directly by
sdelayer -o alter -E calc or
sdelayer -o alter -E minx,miny,maxx,maxy, or indirectly through ArcObjects -- the Feature Extent tab of Layer Properties was added for this purpose). I've only rarely modified the SDE.LAYERS table directly via SQL, but I really can't recommend this for anyone with less than twenty years of SDE/ArcSDE experience (which is probably me and three or four others on the whole planet, and I haven't found it necessary in the past ten years)
Best practice is to initialize the extent at layer creation, or to update it if layer contents change significantly. Of course, if layer contents change significantly, then rebuilding the spatial index is probably warranted as well.
There are number of risks to using SQL to insert/update/delete rows outside the ken of ArcGIS, but they mostly pertain to versioned feature classes, or feature datasets, or the presence CAD objects, or specialized behaviors. Simple feature classes (plain old ArcSDE layers) shouldn't be able to be harmed, though things might get confusing if no features have previously been added to the layer and the spatial index is set to zero (auto-populate), and you don't rebuild the spatial index when you're done inserting.