Fix output correctness issues found in review

[arash77]

The fixes for the points from my review on bgruening#1947, as discussed there.

Output correctness

• Read the input as UTF-8 bytes instead of Path.read_text(), which silently rewrote CRLF/CR line endings and shifted the start indices.
• TSV: escape \, \t, \r and \n and write the row without the csv module, so each chunk stays on one row and the escaping can be reversed. csv.writer additionally quoted every chunk containing a ".
• Write all outputs as UTF-8 instead of the system locale encoding.

Errors

• Clear message for invalid UTF-8 input, for a disallowed special token on the character splitter, and for unsupported or out-of-range custom separator escapes.
• Reject input that is empty or whitespace only instead of writing empty outputs.
• Pass a custom separator as --separator=VALUE, so a separator starting with a dash (e.g. ---) is not read as an option.

Output format

• start_index is null when the splitter returns an invalid position, instead of replacing the number with a message string. The warning on stdout is unchanged and still carries the value.
• One length key per chunk together with the existing length_unit, instead of character_count/token_count.
• strip_whitespace is reported as false in token mode, where the splitter never applies it.

Tests and docs

• New tests: whitespace-only input, disallowed special token on the character splitter, dash-prefixed custom separator, predefined separator inside the repeat.
• The two overlapping token tests now expect start_index: null and the warning, instead of the not_has_text check that only passed because the old code wrote a string there. They stay green now and will fail once the upstream bug (TokenTextSplitter start indices are sometimes -1 langchain-ai/langchain#29884) is fixed, so nothing has to be re-enabled by hand.
• Help documents the tiktoken encoding cache, the TSV escaping and the JSON fields.
• The two tiktoken command line blocks are merged into a macro.

planemo lint clean. planemo test 16/16 before, 20/20 after.