parser: clarify that multiline parser does not join separate events

[Watson1978]

Closes: #338

MultilineParser#parse in lib/fluent/plugin/parser_multiline.rb applies the regexp built from formatN (compiled with Regexp::MULTILINE) to the text it is given, then loops over Regexp::MatchData#post_match and yields once per match. So a single call can emit several records, but every bit of state is local to that call: the plugin keeps no buffer across calls, and therefore cannot combine lines which arrive as separate events. It only exposes has_firstline? / firstline? as helpers for the caller.

The actual line buffering is in lib/fluent/plugin/in_tail.rb: it detects the multiline mode with @multiline_mode = parser_config["@type"].include?("multiline"), and parse_multilines appends lines to a buffer until @parser.firstline?(line) matches the next first line.

lib/fluent/plugin/filter_parser.rb has no guard against multiline: no ConfigError, no warning. So a <filter> with @type parser and @type multiline in its <parse> section silently parses a single field of an already-split event, which is not what users expect. That is the confusion reported in #338. On top of that, filter_one_record adds a new event per yield, so one input event can even be expanded into several events.

The existing sentence only talked about input plugins, so it did not cover the <filter> case at all.

Note that the wording deliberately avoids saying "multiline does not work in filter". The regexp is compiled in multiline mode, so it can still match if the target field value already contains newline characters. What is impossible is joining separate events into one record — hence the pointer to fluent-plugin-concat.