Improve eBPF context propagation stability

[RonFed]

The main goal of this PR is to improve the context propagation in the eBPF code by making it more stable across kernel versions and across instrumented libraries. Those fixes are the result of investigation of Keyval's customers running go auto instrumentation in production.
The changes can be divided into a few categories:

1. Memory allocation: we use bpd_probe_write_user helper function in order to modify user memory. The main challenge is to write to memory which we "allocate" since page faults won't get handled if they occur while the eBPF code attempts to write to such addresses. In order to reduce the page faults I added the following changes:

• madvise and mlock syscalls after we perform the mmap.
• allocate 8 pages per CPU. This number is a result of some experiments I did but the tradeoff as I understand it is: on one side we want to have enough memory available for handling a large number of instrumneted functions without overriding exiting trace memory. On the other side, allocating too much memory increases the chance of page faults - which will cause the write operations to fail, and damage the context propagation.
• Since the added code, I added allocate.go

2. Improving the resiliancy of context propagation relevant functions in the eBPF code. This mainly include error handling and a unified API for all the uprobes.
3. Extracting the Go context key via a new util function.
4. Adding utils function in the eBPF code for reading user memory strings.
5. Change the http server uprobe name to better fit the instrumented function

[edeNFed]

Great work! The code looks much better now. I left a small question regarding stop_tracking_span