Reduce testcase load time

[fjarri]

Currently Jute has to load all the test files and create all the testcase objects before it can proceed with filtering. This means that even if a user only runs a bunch of testcases (which would be a common case during development; the full test suite is run on a CI server, or at least rarely), all the testcase objects still have to be created, producing significant overhead.

One possible way to reduce it is to only do the bare minimum necessary for testcase filtering; everything else, like calculating leaf dependencies or checking the number of function parameters (issue #10) can be done later.

Anything else? How serious will this issue actually be in huge test suites?