This is a cleaner solution for the fix in #438 thanks to the fact that Go 1.19 now is the default and the atomic.Int64 types are automatically aligned correctly on 32 bit systems.

Using this also means that xsync.Int64 can be removed. The new atomic.Int64 type solves the issue and should be quite a lot faster as it avoids the interface conversion.

On some 32-bit architectures, 64-bit atomic operations panic when the
value is not aligned properly.

In this package, this causes netConn operations to panic when compiling
with GOARCH=386, since netConn does atomic operations with int64 values
in the netConn struct (namely, with readExpired and writeExpired).

This commit fixes this by moving readExpired and writeExpired to the
beginning of the struct, which makes them properly aligned.

Glad no one ran into this in production.

Closes #367

Updates #255

Closes #245

Closes #243

NetConn has to close the connection to interrupt in progress reads
and writes. However, it can block reads and writes that occur
after the deadline instead of closing the connection.

Closes #228

See https://github.com/nhooyr/websocket/issues/182

Closes #182

- For JS we ensure we indicate which size initiated the close first from our POV
- For normal Go, concurrent closes block until the first one succeeds instead of returning early

I changed my mind after #103 as browsers include a wasClean event to indicate
whether the connection was closed cleanly. From my tests, if a server using
this library prior to this commit initiates the close handshake, wasClean
will be false for the browser as the connection was closed before it could
respond with a close frame. Thus, I believe it's necessary to fully implement
the close handshake.

@stephenyama You'll enjoy this.

math/rand was being used inappropiately and did not have
a init function for every file it was used in.

Closes #150

- Also allows SetReadLimit to be called concurrently which is a nice
  touch

Closes #153

Realized I can at least make the Reader/Writer/SetReadLimit methods
work as expected even if they're not perfect.

Updates #88

Closes #121

The boys in the office hated it.

Highly unfortunate but makes sense as it seems confusing.

Will be cleaner in GitHub as the README.md is the first
thing you will see instead of a massive directory listing.

Closes #134

Closes #125

Minor API change, will need to document clearly in release notes.

Also simplified and refactored the Conn tests.

More changes soon.

Closes #113

Closes #111
Closes #112