perf(fix): Use boost vector as underlying storage for the buffer instead of the std::vector

[art-gs]

Buffer::prepare() is called every time when sth is about to be written to the buffer, it includes calling of vector::resize().

It turned out that resize() is not so cheap if current capacity is more than requested size, because it touches newly requested chunk of memory.
Measurements showed that boost::vector does it a bit more efficient by using memset() while std::vector uses std::__uninitialized_default_n_a()

SDB-6760

[art-gs]

std vector

% re-lpfixgw-bench --filter=citi_esp
NAME                                                   COUNT       MIN       %50       %95       %99     %99.9    %99.99
------------------------------------------------------------------------------------------------------------------------
citi_esp                                              425630      5.14     5.964     8.388    10.337    45.714   766.135

boost vector

% re-lpfixgw-bench --filter=citi_esp         
NAME                                                   COUNT       MIN       %50       %95       %99     %99.9    %99.99
------------------------------------------------------------------------------------------------------------------------
citi_esp                                              445290     4.526      5.58       8.5     9.981    45.875   1281.28

Note: 99.99% percentage looks worse, it is because boost::vector has different grow factor during reallocation, i.e. it reallocates more often bit using lesser chunks. It is one time thing - buffer grows while the first message is being formatted, subsequent messages re-use the buffer.