Yes, there is. A relevant general framework is the following: for any
2-monad T, we can define notions of pseudo and lax morphism between
T-algebras, and there is a forgetful functor from the 2-category of
T-algebras and pseudo morphisms to the 2-category of T-algebras and
lax morphisms. If T is well-behaved, this forgetful functor has a
left adjoint; see for instance this paper.

There is a 2-monad on the 2-category of Cat-graphs whose algebras are
bicategories, whose pseudo morphisms are pseudofunctors, and whose lax
morphisms are lax functors. Therefore, the above applies to
bicategories. If you trace through the construction, you'll see that
it is given essentially by the recipe you proposed. (This case of the construction can probably be found elsewhere in the literature as well, in more explicit form, but this is the way I prefer to think about it.)

The caveat is that the 2-cells in the 2-categories
defined above are not any of the the usual sort of transformations
between bicategories, only the
icons. (This is what allows you to
have a 2-category containing lax functors.) However, the usual sorts of
transformations are "corepresentable," that is, for any bicategory D
there is a bicategory Cyl(D) such that pseudo or lax functors into
Cyl(D) are the same as pairs of pseudo or lax functors into D and a
pseudo (or lax, with a different definition of Cyl) natural
transformation between them, and likewise we have 2Cyl(D) for
modifications. I believe one can use this to show that in this case,
the construction coming from 2-monad theory does have the property you
want.

Of course, by Chris' question, it seems that this version of L cannot
itself be described as a left adjoint, since there is no 2- or 3-category
containing lax functors and arbitrary pseudo/lax transformations.