#2097 Tika ParserBolt: limit the extracted text with parser.tika.text.maxlength (#2174)

ParserBolt built its text handler with BodyContentHandler(-1), which
removes the limit Tika applies by default, and there was no way to set
one. The whole text of a document was held as a single String next to
the original bytes until the tuple was emitted, so a large body reaching
the bolt could take a lot of the worker heap.

The new key parser.tika.text.maxlength sets the maximum number of
characters of text kept per document. When the limit is reached Tika
stops the parse with a WriteLimitReachedException; the bolt now keeps
the text and outlinks extracted up to that point, sets
parse.text.trimmed to true in the metadata, counts it under
text_trimmed and emits the document instead of failing it with a parse
error. The default stays -1 so the behaviour does not change unless the
key is set; any negative value means no limit, since Tika itself only
accepts exactly -1 and fails every document on other negative values.

The parse timeout described in the issue is left for a follow-up.