enhancement(buffers): Instrument buffer total bytes/events received, sent, dropped (#9327)

* Add internal_event dependency

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Add EventsReceived, Sent, Dropped skeletons

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Implement emit_metrics for relevant InternalEvents

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Instrument EventsSent for in-memory buffer

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Instrument EventsSent for disk buffer

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Instrument EventsDropped

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Addd instrumentation wrapper for blocking buffer

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Fix clippy errors

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Update cue data for docs

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Update dependencies

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Use encoded byte size for disk buffer EventsSent

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Emit EventsReceived in writer for disk buffer

We emit in try_send which calculates event size to avoid an expensive recalculation

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Add ByteSizeOf trait bound for calculating byte size for memory buffer events

Rewrites comment for BlockWhenFull. Refactor DropWhenFull start_send

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Use appropriate byte size for blocking buffers

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Refactor instrumentation wrapper to avoid repetition

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Fix type errors

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Remove unnecessary comment

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Fix formatting

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Rename events with buffer prefix

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Update sink telemetry docs

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Update transform telemetry docs

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Remove transform telemetry cue update

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Add tracing span to buffer variant struct

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Use span for correct tagging in memory buffer buffer_received_ metrics

Also replaces Option<Span> with the notion of enabled/disabled spans.
The disabled span serves as a flag for whether or not to instrument a
memory buffer (not in the case of buffers attached to transforms).

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Disable instrumentation for memory buffers when span is disabled

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Use span for correct tagging in disk buffer buffer_received_ metrics

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Remove emit_log in impl

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Replace allocated_bytes with size_of

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Refactor span usage for buffer instrumentation to fix errors

Instead of passing spans in as part of buffer variants, we pass them
as additional arguments in the buffer build function. This simplifies
the wiring and better separates concerns for the variant type.

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Add DebuggingRecorder to buffer benchmarks

To properly benchmark instrumented buffers, a metrics recorder needs to be
installed. Otherwise, metrics related logic is a noop and performance overhead
is not accurately reflected.

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Add metrics tracing layer dependencies to buffer benchmarking

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Use flag for disabling instrumentation in transform memory buffers

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Use shared state and tokio task for metrics emission

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Update tokio runtime in buffer benchmarks

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Organize code and fix linting issues

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Remove unnecessary file

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Move init_instrumentation into new for buffer usage data

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Use u64 instead of usize when possible

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Use batched instrumentation mechanism for disk buffers

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Use swap instead of load to avoid overcounting metrics

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Add buffer_events, buffer_byte_size gauge

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Resolve clippy errors for cast precision loss (allow)

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Add buffer_events and buffer_byte_size to cue docs

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Combine event and byte size increment functions

Signed-off-by: 001wwang <will.wang@datadoghq.com>

* Rename _bytes metrics to _event_bytes to match component style

Signed-off-by: 001wwang <will.wang@datadoghq.com>