Add TCO to Lisp

[vinc]

Add simple support for Tail Call Optimization by following how it is done on Lispy: http://norvig.com/lispy2.html

[vinc]

Before a deep recursion would lead to a stack overflow:

> time lisp /tmp/lisp/sum.lsp 100
5050
Executed 'lisp /tmp/lisp/sum.lsp 100' in 0.020997s

> time lisp /tmp/lisp/sum.lsp 1000
500500
Executed 'lisp /tmp/lisp/sum.lsp 1000' in 0.052992s

> time lisp /tmp/lisp/sum.lsp 10000
Error: Could not allocate address 0xffffffffe8

Now deep recursions are possible but everything is a bit slower:

> time lisp /tmp/lisp/sum.lsp 100
5050
Executed 'lisp /tmp/lisp/sum.lsp 100' in 0.023996s

> time lisp /tmp/lisp/sum.lsp 1000
500500
Executed 'lisp /tmp/lisp/sum.lsp 1000' in 0.124981s

> time lisp /tmp/lisp/sum.lsp 10000
50005000
Executed 'lisp /tmp/lisp/sum.lsp 10000' in 28.046145s

[vinc]

Another speed comparison, before:

> time lisp /tmp/lisp/pi.lsp 1000
3.1415926535897932384626433832795028841971693993751058209749445923078164062862089986280348253421170679821480865132823066470938446095505822317253594081284811174502841027019385211055596446229489549303819644288109756659334461284756482337867831652712019091456485669234603486104543266482133936072602491412737245870066063155881748815209209628292540917153643678925903600113305305488204665213841469519415116094330572703657595919530921861173819326117931051185480744623799627495673518857527248912279381830119491298336733624406566430860213949463952247371907021798609437027705392171762931767523846748184676694051320005681271452635608277857713427577896091736371787214684409012249534301465495853710507922796892589235420199561121290219608640344181598136297747713099605187072113499999983729780499510597317328160963185950244594553469083026425223082533446850352619311881710100031378387528865875332083814206171776691473035982534904287554687311595628638823537875937519577818577805321712268066130019278766111959092164201989
Executed 'lisp /tmp/lisp/pi.lsp 1000' in 0.180972s

After:

> time lisp /tmp/lisp/pi.lsp 1000
3.1415926535897932384626433832795028841971693993751058209749445923078164062862089986280348253421170679821480865132823066470938446095505822317253594081284811174502841027019385211055596446229489549303819644288109756659334461284756482337867831652712019091456485669234603486104543266482133936072602491412737245870066063155881748815209209628292540917153643678925903600113305305488204665213841469519415116094330572703657595919530921861173819326117931051185480744623799627495673518857527248912279381830119491298336733624406566430860213949463952247371907021798609437027705392171762931767523846748184676694051320005681271452635608277857713427577896091736371787214684409012249534301465495853710507922796892589235420199561121290219608640344181598136297747713099605187072113499999983729780499510597317328160963185950244594553469083026425223082533446850352619311881710100031378387528865875332083814206171776691473035982534904287554687311595628638823537875937519577818577805321712268066130019278766111959092164201989
Executed 'lisp /tmp/lisp/pi.lsp 1000' in 0.627904s

[vinc]

Removing the initial cloning of exp improved the perf a little bit:

> time lisp /tmp/lisp/pi.lsp 1000
3.1415926535897932384626433832795028841971693993751058209749445923078164062862089986280348253421170679821480865132823066470938446095505822317253594081284811174502841027019385211055596446229489549303819644288109756659334461284756482337867831652712019091456485669234603486104543266482133936072602491412737245870066063155881748815209209628292540917153643678925903600113305305488204665213841469519415116094330572703657595919530921861173819326117931051185480744623799627495673518857527248912279381830119491298336733624406566430860213949463952247371907021798609437027705392171762931767523846748184676694051320005681271452635608277857713427577896091736371787214684409012249534301465495853710507922796892589235420199561121290219608640344181598136297747713099605187072113499999983729780499510597317328160963185950244594553469083026425223082533446850352619311881710100031378387528865875332083814206171776691473035982534904287554687311595628638823537875937519577818577805321712268066130019278766111959092164201989
Executed 'lisp /tmp/lisp/pi.lsp 1000' in 0.470081s

[vinc]

Removing another cloning improved the performance a lot:

> time lisp /tmp/lisp/pi.lsp 1000
3.1415926535897932384626433832795028841971693993751058209749445923078164062862089986280348253421170679821480865132823066470938446095505822317253594081284811174502841027019385211055596446229489549303819644288109756659334461284756482337867831652712019091456485669234603486104543266482133936072602491412737245870066063155881748815209209628292540917153643678925903600113305305488204665213841469519415116094330572703657595919530921861173819326117931051185480744623799627495673518857527248912279381830119491298336733624406566430860213949463952247371907021798609437027705392171762931767523846748184676694051320005681271452635608277857713427577896091736371787214684409012249534301465495853710507922796892589235420199561121290219608640344181598136297747713099605187072113499999983729780499510597317328160963185950244594553469083026425223082533446850352619311881710100031378387528865875332083814206171776691473035982534904287554687311595628638823537875937519577818577805321712268066130019278766111959092164201989
Executed 'lisp /tmp/lisp/pi.lsp 1000' in 0.221966s