Previously CodeMetaInfoParser saved MatchResults for each meta info in
text (and for each meta info it was different string with text because
of extracting parsed part from original text), so it led to O(n^2)
memory complexity (where n is a size of original text).
Such complexity caused OOME in case of big testdata (~1k meta infos
for text of 4k lines)
This is needed to provide ability for declaring new implementations of
test generators, based on existing infrastructure, which won't add
dependency on :compiler:tests-common
Also this commit removes implicit dependency on :compiler:tests-common
from :compiler:tests-common-new
All of new classes lays in lays in :compiler:tests-common-new module
which includes classes for FE 1.0 and FIR diagnostics tests and
JVM black boxtests
It contains different abstractions which represents parts of compiler
pipeline and artifacts produced by them, service structure, handlers
for analysis of artifacts
Dmitriy Novozhilov