gdb_server: fix GDB_BUFFER_SIZE usage, fix unaligned access during bulk transfers
Currently size of the GDB buffer is 16384 bytes but it is treated as
nul-terminated string in most of the code, so effective size of the
buffer is actually 16383 bytes. OpenOCD responds with `PacketSize=3fff`
to qSupported request. Result of GDB's `m` command is encoded in hex so
each data byte uses two bytes in the buffer. As a result GDB will split
bulk read requests into chunks 0x1fff bytes each. This causes troubles
on targets (or memory regions) which support only aligned, word-sized
access (such as MMIO buffers).
OpenOCD:
Error: Failed to read memory at 0x40321ffe
Error: Failed to read memory at 0x40321000
Error: Failed to read memory at 0x40323000
Error: Failed to read memory at 0x40325ffe
Error: Failed to read memory at 0x40329ffa
Error: Failed to read memory at 0x40329ffc
Error: Failed to read memory at 0x4032bffc
Error: Failed to read memory at 0x4032dffa
Consolidate GDB_BUFFER_SIZE usage: ensure size of each buffer is
(GDB_BUFFER_SIZE + 1), add explicit comment that additional byte is used
for nul-termination. Report correct size of the buffer to GDB (0x4000)
as recommended in GDB's docummentation: `if the stub stores packets in a
NUL-terminated format, it should allow an extra byte in its buffer for
the NUL`
Checked with clang-asan, clang-analyzer, valgrind - no new errors.