MarkdownV2Parser.parse ignores backslash escapes and deviates from the spec in several places

[zeynalnia]

Problem

MarkdownV2Parser.parse (in gramjs/extensions/markdownv2.ts) departs from
the Telegram MarkdownV2 spec
in several ways:

1. Backslash escapes are not honored. Per the spec, \X for any X in
   _*[]()~`>#+-=|{}.! becomes the literal X, and \\ becomes \. Today
   these are passed through verbatim:

   • Input 1\.5 → output text 1\.5 (expected 1.5).
   • Input \*not bold\* → output text <b>not bold</b> (expected literal
     text *not bold* with no entity, since the delimiters are escaped).

2. Italic uses - instead of _. The current code matches -text- for
   italic. The spec uses _text_ and reserves __text__ for underline.

3. No blockquote support. Per spec, lines beginning with > form a
   blockquote, and a final line ending in || marks it as expandable
   (MessageEntityBlockquote.collapsed = true). Today these are emitted as
   literal > chars.

4. Per-region escape rules are not applied. Inside pre and code only
   \\ and \` should unescape; inside the (URL) of a link or custom
   emoji only \\ and \) should unescape. There's no pass to apply this
   selectively.

5. HTML special characters in plain text confuse the downstream HTML
   parser. A user-typed < is not escaped before being handed to
   HTMLParser.parse, so <b>not bold</b> typed into MarkdownV2 input is
   incorrectly interpreted as bold.

6. HTMLParser is missing some Telegram HTML-spec tags that
   htmlToMarkdownV2 (and external callers) need to round-trip cleanly:
   <tg-spoiler>, <span class="tg-spoiler">, <ins> (underline
   alternative), <strike> (strikethrough alternative). And
   HTMLParser.unparse emits the library-internal <spoiler> tag (not in
   the spec) and drops the expandable attribute on collapsed blockquotes,
   so the flag doesn't survive round-trips.

Proposal

Rewrite the markdown→HTML transform inside the existing
markdown → HTML → HTMLParser pipeline as a staged process: extract
protected regions (pre/code/link/emoji) up front with their own escape
rules; mask remaining backslash-escapes; HTML-escape & and < in user
content; run span and blockquote markup; unmask; restore protected regions.
Switch italic to _ per spec. Expose markdownV2ToHtml and
htmlToMarkdownV2 as standalone functions so external callers can convert
between formats. Patch HTMLParser to recognize the missing tag forms and
to preserve the expandable attribute on round-trip.