* explain why tcb_t is added at the end of tls allocation

* fix calculation of TLS size for main fibril

Before this patch, _tcb_data_offset always used __progsymbols.elfstart. However,
that is wrong when it is being called from the loader server!! Now we pass to it
a pointer to the correct ELF, falling back to elfstart in the public tls_get call.

(debugging this was quite an ordeal and took me like 5 hours, thanks for asking (':)

* a few comments for TLS allocation

Devman will only return value when the entire device tree is stable.

Now it is guaranteed that, if the volume was inserted at power-on time,
it is mounted at this point. This is because (1) srv_start(devman)
only returns once all devices have been enumerated and
(2) srv_start(volsrv) only returns once all volumes have been mounted.

To determine if legacy IDE I/O ports are free, isa-ide asks isa,
who asks pciintel. pciintel waits for bus enumeration to complete,
then waits for all functions except the one who is asking
(which is ISA bus) to stabilize. During attach pci-ide will claim
the legacy IDE ports. Thus, if at this point legacy IDE ports
are unclaimed, pciintel tells ISA they are free, which tells isa-ide,
which continues to attach. If they are not free, isa-ide will not
attach.

This works for all use cases, including system without PCI bus,
system with PCI bus, but no (or disabled) PCI IDE, system with PCI
IDE with unrecognized VID/PID (which we will handle in legacy ISA mode).

…elenOS#242)

init RTLD runtime at load time even for statically linked binaries