Removed the duplicated function body in the Cortex-M4 AC6 port

[fdesbiens]

Fixes #569, reported by @AKevin99.

The defect

_tx_thread_system_return_inline() in ports/cortex_m4/ac6/inc/tx_port.h was followed by a second, orphaned copy of its own body:

        __restore_interrupt(interrupt_save);
    }
}
unsigned int interrupt_save;          /* <- no function header above this */

    /* Set PendSV to invoke ThreadX scheduler.  */
    *((volatile ULONG *) 0xE000ED04) = ((ULONG) 0x10000000);
    ...

The copy has no function header, so it declares interrupt_save at file scope and then places statements there. That does not compile:

error: expected identifier or '(' before 'volatile'
    *((volatile ULONG *) 0xE000ED04) = ((ULONG) 0x10000000);
       ^~~~~~~~

It sits inside #if defined(__GNUC__) || defined(__ICCARM__), which ARM Compiler 6 satisfies, and outside TX_DISABLE_INLINE, so it is reached by a default AC6 build of this port.

The orphan is also the older text: it lacks the dsb and isb barriers that the live copy above it has. Those barriers were added by 33efad3 (#523), and the previous version was not deleted, which matches the "small oversight when improving the code" in the report.

The fix

Delete the orphaned copy. Nothing else changes.

What remains is exactly the body every sibling carries. After this change, ports/cortex_m4/ac6/inc/tx_port.h differs from ports/cortex_m4/iar/inc/tx_port.h and from ports/cortex_m7/ac6/inc/tx_port.h only in the port name in the banner and the version string.

Worth stating explicitly: the surviving body is correct as it stands and needs no barrier added. Diffing AC6 against the GNU sibling suggests otherwise at first glance, because the orphan lines up against a trailing isb that GNU has, but grouping all eighteen M3/M4/M7 headers by their actual function body shows AC6 belongs with IAR, which has no trailing isb. Adding one here would have been a real behavioural change smuggled in under a cleanup.

Verification

• The function compiles cleanly in isolation after the change, and fails on the file scope statements before it.
• All 208 tx_port.h files in the repository were scanned structurally, tracking brace depth and ignoring preprocessor lines, for a PendSV write occurring outside any function body. This file was the only hit, both before and after.

Related divergence, not addressed here

Grouping the eighteen M3/M4/M7 headers by that function body turns up three distinct versions of code that scripts/copy_armv7_m.sh describes as coming from a single source:

Body	Files
dsb+isb at entry, no trailing isb	ac6 and iar for M3, M4, M7
dsb+isb at entry, plus a trailing isb after __restore_interrupt()	gnu for M3, M4, M7
No barriers at all	ac5 and keil for M3, M4, M7, and ports_arch/ARMv7-M/threadx/inc/tx_port.h

The last row is the one to look at. scripts/copy_armv7_m.sh says "There is only one tx_port.h file that covers three architectures: M3/M4/M7 and four tools: ac5/ac6/gnu/iar", and the file it copies from is in the no-barrier group. Running that script today would therefore revert #523 across every M3, M4 and M7 port that currently has the barriers.

That is out of scope for a fix to #569. I will address that in one or more PRs, however.