refactor(http-clients)!: Replace crawl() and stream() with send_request - #2349

…on reuse (#2373)

Two tests added in #2349 fail intermittently in CI (seen on #2371).

- `test_closing_unread_response_reuses_connection[impit]`: impit returns
a drained connection to its pool in the background, so the request sent
right after a close sometimes opens a second connection. Under load this
failed in 2 of 120 local runs. The test now allows up to two
connections. A working client opens 1 or 2, and one that drops the
connections of closed responses always opens 3 (300 runs each), so the
test still catches that regression.
- `test_read_again_after_timeout[httpx]`: the 200 ms timeout covered the
whole request, and the slow macOS runner didn't get the headers in time.
The test now uses a 10 s timeout and moves the event loop clock past the
deadline before the second `read()`, so it doesn't wait on the wall
clock.

Both tests still fail when the behavior they cover is broken. After the
change, 1,440 runs under load passed.

*✍️ Drafted by Claude Code*

…HttpClient.crawl` (#2367)

### Description

`HttpCrawlingResult` is removed. Only `HttpClient.crawl` returned it,
and #2349 removes `crawl`. `HttpCrawlingContext` used to inherit
`http_response` from it and now declares the field itself, so the
crawling contexts work as before. The v2 upgrade guide notes the
removal.

### Issues

- Follow-up: #2349

*✍️ Drafted by Claude Code*

### Description

A `Cookie` header passed with a request or set on the client is now
merged with the session cookies, with every HTTP client and in
`context.send_request` of `PlaywrightCrawler`. A cookie in the header
wins over a session cookie of the same name and isn't stored in the
session. An empty `Cookie` header no longer keeps the session cookies
out. The v2 upgrade guide describes the change.

### Issues

- Part of: #2307
- Follow-up: #2349

*✍️ Drafted by Claude Code*

Co-authored-by: Vlada Dusek <v.dusek96@gmail.com>