Fix Python 2 syntax in cg, common, lr, nb, regression

Summary

cg.py, common.py, lr.py, nb.py, and regression.py were written for Python 2 and currently fail to import on Python 3 with SyntaxError (or TypeError: unsupported operand type(s) for >> at runtime once parsing is past). Earlier branches fix/ml-cg-syntax and fix/cg-exception-style only updated the except Exception, e: syntax and did not touch the print >> sys.stderr, ... and print 'literal' statements, the raise NameError, ('error') statement, or the Python 2 / integer-division in common.plot_sequence_data. Without this fix the only way to run python3 cg.py is to run an older Python; the README and most users today run a recent Python where the files simply don't import.

Changes

cg.py:

• 7× print >> sys.stderr, X → print(X, file=sys.stderr) (the inner loop's iteration counter line, the exception trace line, the line-search line, the trailing empty line).
• 1× raise NameError, ('error') → raise NameError('error') (the NaN/Inf guard).
• 3× except Exception, e: → except Exception as e: (line-search, secant/bracket steps, and the inner except that bisects w3 after a NaN/Inf step).

common.py:

• 6× print >> sys.stderr, X → print(X, file=sys.stderr) in plot_sequence_data (the wide-row branch's three lines plus the narrow-row branch's three lines).
• 3× d / 2 / (L - 1) / 2 / len(s_line) / w → d // 2 / (L - 1) // 2 / len(s_line) // w so the multiplication counts come out as integers — Python 2 / on two ints is floor-div, Python 3 / is true division; without //, ' ' * 1.5 raises TypeError: can't multiply sequence by non-int of type 'float' the first time the function is called with two rows of different length.

nb.py:

• 8× print >> sys.stderr, X → print(X, file=sys.stderr) (the parameter-error guard, the top-features header and rows, the four training/test accuracy lines).

lr.py:

• 1× print 'Done with function evalution C = %d' % self.c → print('Done with function evalution C = %d' % self.c) (this is the bare print statement that the SyntaxError was pointing at; the print >> lines below were already valid for Python 2 by accident but I've also normalized them to print(..., file=sys.stderr) for consistency).
• 2× print >> sys.stderr, '... accuracy ...' → print('... accuracy ...', file=sys.stderr).

regression.py:

• 1× print 'Done with function evalution C = %d' % self.c → print(...) (the same SyntaxError source as in lr.py).
• 1× print >> sys.stderr, 'Test RMSE : %lf' % rmse → print('Test RMSE : %lf' % rmse, file=sys.stderr).

Verification

• python3 -c "import ast, sys; ast.parse(open(p).read())" passes for all five files (and all others: adaboost.py, cf.py, dt.py, gda.py, hmm.py, mf.py, nn.py, perceptron.py, softmax.py, svm.py).
• python3 -c "import cg, common; common.trace()" runs without error and prints function = <module> , line = 1 (was raising TypeError from the print >> inside trace before the prior fix/common-trace-python3-print-and-narrow-except branch).
• python3 -c "import common; common.plot_sequence_data(['abc', 'de'], ['12', '34'], w=8)" runs to completion and prints the alignment visualization (was raising TypeError: can't multiply sequence by non-int of type 'float' because m = d / 2 was 0.5 instead of 0).
• Line endings preserved: lr.py and regression.py retain their original CRLF line terminators; the executable bit on common.py is preserved. git diff --stat shows 5 files changed, 37 insertions(+), 37 deletions(-) — the symmetric line count is the signature of mechanical 1:1 syntax conversions, not a refactor.
• The other algorithmic files (adaboost, cf, dt, gda, hmm, mf, nn, perceptron, softmax, svm) all still have Python 2 print >> statements inside function bodies; those only fail at runtime when the print is actually reached, so they don't surface as SyntaxError at import time. They are out of scope for this change.