Replace regexp tokenizer with recursion parser

In the current implementation, the RegexTokenizer uses regular
expressions to tokenize the format string. However, regular expressions
are not well-suited for parsing nested structures like nested brackets
because they cannot match patterns with recursive depth.

To address this, we'll write a custom recursive parser that can handle
nested optional brackets. This parser will process the format string
character by character, building the format items and managing the
nesting of optional components.

That change allowed me to cover `strict_date_optional_time` which has a
nested depth according to the examples in [OS repo][2]

Additional tests taken from [the doc][1]

[1]: https://opensearch.org/docs/latest/field-types/supported-field-types/date/
[2]: https://github.com/opensearch-project/OpenSearch/blob/main/server/src/test/java/org/opensearch/common/time/DateFormattersTests.java

quickwit/quickwit-datetime/Cargo.toml

@@ -13,7 +13,6 @@ license.workspace = true
 [dependencies]
 anyhow = { workspace = true }
 itertools = { workspace = true }
-regex = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }
 tantivy = { workspace = true }