[wasm-ep] Minimal diagnostic tracing configuration and sample (#69158)

[wasm-ep] Minimal diagnostic tracing configuration and sample (#69158)

* Adds a `/p:WasmEnablePerfTracing=true` configuration.

   In this configuration the runtime is built with threading (`MonoWasmThreads` property is true), but user C# code is not allowed to start threads and doesn't use the portable threadpool (`MonoWasmThreadsNoUser` property is also true).

   The upshot is that EventPipe can start threads but user code is still single-threaded.

* Adds a `MONO.diagnostics` interface in JavaScript.  There's a single method for now `createEventPipeSession` which creates a session that can save a trace to a file on the virtual file system.  JS code (or a user in a dev console) needs to call `start()` and `stop()` on the session object to begin collecting samples.    The data is saved temporarily to the Emscripten VFS and can be retrived into a JavaScript Blob (and from there downloaded to a file outside the browser).

* Adds a sample that runs an async task for five seconds and collects samples and then triggers a `click()` to download the trace file out of the browser.

* Adds a TS module to help with working with uint64_t values in the emscripten heap.

* Exposes empscripten Module.stackSave, stackRestore and stackAlloc operations to the runtime TS modules.  Use for working with event pipe session ID outparam.

---

* add DISABLE_WASM_USER_THREADS mono cmake option

* Disable Thread.StartInternal icall if DISABLE_WASM_USER_THREADS

   if threading is enabled for the runtime internally, but disabled for user code, throw PNSE

* add an eventpipe sample

* [wasm-ep] (browser-eventpipe sample) run loop for longer

* [samples/wasm-eventpipe] make an async task sample

   change the sample to do some work asynchronously using setTimeout instead of blocking

* [wasm] Add MONO.diagnostics interface

   Binds enable, start, disable methods defaulting to non-streaming FILE mode

* if wasm threads are disabled, but perftracing is enabled, don't log overlapped io events

* fix whitespace and nits

* don't need try/finally in the sample anymore

* more whitespace

* add start method to EventPipeSession interface

* don't run wasm-eventpipe sample on CI lanes without perftracing

* more whitespace

* fix eslint warnings, default rundown to true, allow callback for traceFilePath option

* add EventPipeSession.getTraceBlob

   for retrieving the collected traces instead of exposing the emscripten VFS directly.

   update the sample to use URL.createObjectURL (session.getTraceBlob()) to create the download link

* [browser-eventpipe sample] remove unnecessary ref assemblies

* use ep_char8_t for C decls of event pipe wasm exports

* Use stack allocation for temporaries

   Expose the emscripten stack allocation API

* Use 32-bit EventPipe session ID on WASM

   64 bit integers are awkward to work with in JavaScript.

   The EventPipe session ID is derived from a pointer address, so even though it is nominally a 64-bit value, in practice the top bits are zero.

   Use a 32-bit int to represent the session ID on the javascript side and convert to 64-bit in C when calling down to the EventPipe APIs

* Make the sample do more work in managed

   give the sample profiler some non-empty samples to collect

* Move withStackAlloc to memory.ts

* simplify VFS .nettrace file naming

   Just use consecutive integers to uniquify the session traces.  Dont' need a fancy timestamp in the VFS (which would also not be unique if you create sessions below the timestamp resolution)

* Add overloads to memory.withStackAlloc to avoid creating closures

   Pass additional arguments to the callback function

* sample: explain why there's a 10s pause

* move createEventPipeSession callback to a function

   ensures the closure is created once

* Use a tuple type for withStackAlloc

* use unsigned 32-bit get/set in cuint64 get/set

* fix whitespace