I'm a bit confused by the difference taken between the base "Lemma" and the MANY "Senses" it could have. Normally, each lemma has a SINGLE sense and a standard/base form. Some lemmas could have several forms (orthographic), while keeping its single sense. Each sense however may have restrictions on its forms depending on context/sentence (e.g. capitalization) but this also applies to the lemma.
May be the term "Lemma" just refers to a dictionary entry which may exhibit several related senses (with minor semantic difference, meaning that the lemma also has the same set of possible translations to the same target language: note however that some lemmas may not have any suitable translation into a single lemma in the target language, where an expression could be needed, and each translation will depend on the forms and context of use or usage where one or several lemmas from the source language may map to one or several target lemmas in the target).
But I don't see how separating senses from lemmas will offer any help, it is in my opinion an extra and unneeded layer of complexification.
Are there good counter examples ? I can't imagine anyone (unless lemmas are ill-defined: if you refer to a dictionnary entry, it is just an editorial choice from a specific dictionnary).
May be this is just an informal group of related senses but it is highly debatable and depend on each author: some authors may create entries by level of language, or jargons/terminologies/context of use (e.g. legal, commercial, vulgar/vernacular, scientific in specific domains), and such grouping is generally evolutive (even from the same authors) and subject to lot of personal perceptions and interpretations...
So we should make things more simple: merge Lemas and Senses into the same entity type (1-to-1). The only difference I see is in the set of forms for the same lemma, which may be euivalent (with just one prefered in some contexts, such as abbreviated forms, slangs/alterations/simplifications or just forms that are always considered as equivalent (e.g. indetermination of accents, proposed orthographic reforms, historic forms that fell out of use...)
Now there's the special case of contextual mutations (generally for phonetics or harmony, including some unwritten parts, such as rules for contractions, elisions and liaisons in French that change how surrounding terms are written or modified outside the written (or spoken, or gestured) form of the lemma itself., or the insertion of non-semantic phonetic particles (like [-t-] or [z'] in French)