feat(#256): add roll-bases lint

[VasilevNStas]

fix #256

What

A new misc/roll-bases lint that reports composite @base attributes
which could be written in a shorter, rolled form.

Consider this XMIR:

<o base=".foo">
  <o base="x"/>
</o>

It can (and must) be written shorter:

<o base="x.foo"/>

The lint reports the unrolled form with a warning.

Why

As @maxonfjvipon correctly pointed out, XMIR is not only produced by
eo-parser (which already rolls bases during parsing) — it may also come
from phino, jeo-maven-plugin, or be written by hand. In those inputs
unrolled composite bases are a real mistake, and nothing currently flags
them.

How it works

The lint follows the collapse rules of roll-bases.xsl from eo-parser:
a node with @base starting with . and a single child whose @base
does not start with ., has no inner objects, no data and no name, must
be merged into concat(child/@base, @base).

Why unit tests instead of yaml packs

The yaml packs always parse EO code, and eo-parser rolls bases on that
path — so an unrolled catches case cannot be expressed in a pack. The
lint is therefore covered by unit tests in LtByXslTest that feed a
hand-built XMIR (via Xembler) straight to LtByXsl:

• catchesUnrolledBases — .foo/x is reported
• allowsRolledBase — x.foo is clean
• allowsNestedBaseWithData — a child with data must not be rolled
• allowsDoubleDotBases — .foo/.bar must not be rolled

Both mvn test (594 tests) and mvn clean install -Pqulice pass.

[VasilevNStas]

@volodya-lombrozo @maxonfjvipon please review this PR