Abilty to set up non-OHMS OH chunks for vector search for AI chat

[jrochkind]

We can't link to these very well, at present just to Work page. But still can form answers on them.

Stil a few left out due to error, but we want to prototype most of em.

OralHistoryContent#id 's : 470 , 131, 577, 127, all produce error: undefined method 'speaker_name' for nil at transcript_chunker.rb:226:in 'OralHistory::TranscriptChunker#build_chunk_record'

---

• extract OralHistoryContent::Paragraph class to represent paragraphs from multiple kinds of transcript sources, for use by chunker
• OralHistory::PlainTextParagraphSplitter
• change name of OhmsLegacyTranscriptChunker to TranscriptChunker because we're going to make it able to handle more things
• refactor transcript chunker to be ready to handle other types of transcripts
• remove fragment and transcript ID from base/plaintext paragraph, actually only relevant to OHMS display
• TranscriptChunker can handle plaintext searchable_transcript_text now
• fix, mutating gsub! does not return receiver
• handle idiosyncratic plain text transcript
• keep going in error condition where there is no speaker name