http server instrumentations: make it possible to sample according to captured headers

I would like to do sampling depending on any captured request header.

This should be doable with the Composite Rule Based Sampler but we need instrumentations to collect these headers before the span is created. AFAICS no http server instrumentation does that. The capturing of headers is opt-in so any eventual performance hit will be paid only if requested.

Instrumentations to fix:

• aiohttp-server, see create_aiohttp_middleware
• django, see _DjangoMiddleware.process_request
• asgi, see OpenTelemetryMiddleware.__call__
• falcon, see _InstrumentedFalconAPI.__call__
• flask, see _wrapped_before_request
• pyramid, see _before_traversal
• tornado, see _start_span
• wsgi, see OpenTelemetryMiddleware.__call__

Start by inspecting the listed entry points: create_aiohttp_middleware, _DjangoMiddleware.process_request, OpenTelemetryMiddleware.call, _InstrumentedFalconAPI.call, _wrapped_before_request, _before_traversal, _start_span, and the WSGI middleware call. Trace when each server span is created and how opt-in captured headers are currently handled. Done means captured request headers are available before span creation in all listed instrumentations so sampling can use them.