update to tesseract 5

[keighrim]

A recent small-scale in-house experiments show that Tesseract 5 (T5)'s outperforms Tesseract 4 in terms of accuracy measure (CER). Hence we would like to update this app to T5 instead of retiring it in favor of apps based on larger models (such as doctr or llava). This will keep an entry in the app directory for a "good-enough" text recognition under restricted HW environment.

I'd also like to take this update work as an opportunity to partially address clamsproject/app-role-filler-binder#4 issue, in that the updated T5 wrapper app should use the current docTR app as a reference implementation to organize input/output MMIF structure. This means, we would like to wrap tesseract's internal structure understanding scheme in our vocabulary terms. Concretely, in docTR-wrapper, we translated

1. Page: no translation as we only deal with single-page scenario (one image at a time)
2. Block --> Paragraph from LAPPS vocab
3. Line --> Sentence from LAPPS vocab
4. Word --> Token from LAPPS vocab

Similarly with T5, we translate

1. Page: no translation as we only deal with single-page scenario (one image at a time)
2. Block --> Paragraph from LAPPS vocab
3. Par: no translation as we already used LAPPS Paragraph, and there's no middle level between Paragraph and Sentence in LAPPS vocab.
4. Line --> Sentence from LAPPS vocab
5. Word --> Token from LAPPS vocab

This will naturally address #18 once implemented.